Merge remote-tracking branch 'origin/litellm_fix_agent_mcp_grants' into litellm_fix_agent_mcp_grants

.circleci/config.yml

@@ -112,10 +112,10 @@ commands:
             node --version
             npm --version
   install_rust:
-    description: "Install pinned rustup (1.28.2) and Rust toolchain (1.97.1) with checksum verification. Adds ~/.cargo/bin to PATH. Run this before any `uv sync` or `uv build` of the workspace: the root package builds litellm-rust through maturin, and on an image without cargo maturin fetches an unpinned rustup and a floating toolchain by itself."
+    description: "Install pinned rustup (1.28.2) and Rust toolchain (1.98.0) with checksum verification. Adds ~/.cargo/bin to PATH. Run this before any `uv sync` or `uv build` of the workspace: the root package builds litellm-rust through maturin, and on an image without cargo maturin fetches an unpinned rustup and a floating toolchain by itself."
     steps:
       - run:
-          name: Install Rust (rustup 1.28.2, toolchain 1.97.1)
+          name: Install Rust (rustup 1.28.2, toolchain 1.98.0)
           command: |
             case "$(uname -m)" in
               x86_64)
@@ -135,7 +135,7 @@ commands:
               "https://static.rust-lang.org/rustup/archive/1.28.2/${RUSTUP_TRIPLE}/rustup-init"
             echo "${RUSTUP_SHA256}  /tmp/rustup-init" | sha256sum -c -
             chmod +x /tmp/rustup-init
-            /tmp/rustup-init -y --no-modify-path --profile minimal --default-toolchain 1.97.1
+            /tmp/rustup-init -y --no-modify-path --profile minimal --default-toolchain 1.98.0
             rm -f /tmp/rustup-init
             echo 'export PATH="$HOME/.cargo/bin:$PATH"' >> "$BASH_ENV"
             export PATH="$HOME/.cargo/bin:$PATH"
@@ -300,7 +300,7 @@ jobs:
             if ($rustupActual -ne $rustupExpected) {
               throw "rustup installer hash mismatch: expected $rustupExpected got $rustupActual"
             }
-            & $rustupInit -y --profile minimal --default-toolchain stable
+            & $rustupInit -y --profile minimal --default-toolchain 1.98.0
             if ($LASTEXITCODE -ne 0) {
               exit $LASTEXITCODE
             }

.github/scripts/smoke_test_native_wheel.py

@@ -0,0 +1,68 @@
+from __future__ import annotations
+
+import subprocess
+import sys
+import tempfile
+import zipfile
+from pathlib import Path
+from typing import Final
+
+CHILD_SCRIPT: Final = """
+from importlib.util import module_from_spec, spec_from_file_location
+from pathlib import Path
+import sys
+
+native_path = Path(sys.argv[1])
+spec = spec_from_file_location("litellm.rust_bridge._native", native_path)
+if spec is None or spec.loader is None:
+    raise RuntimeError("cannot create native extension import specification")
+module = module_from_spec(spec)
+spec.loader.exec_module(module)
+
+before = module.gil_stats()
+if not isinstance(before.get("releases"), int):
+    raise AssertionError(f"unexpected gil_stats result: {before!r}")
+
+try:
+    module._panic_for_test()
+except BaseException as error:
+    if type(error).__name__ != "PanicException":
+        raise AssertionError(f"expected PanicException, got {type(error).__name__}") from error
+else:
+    raise AssertionError("Rust panic returned without raising")
+
+after = module.gil_stats()
+if not isinstance(after.get("releases"), int):
+    raise AssertionError(f"native module unusable after panic: {after!r}")
+"""
+
+
+def main() -> int:
+    if len(sys.argv) != 2:
+        sys.stderr.write(f"usage: {Path(sys.argv[0]).name} WHEEL\n")
+        return 2
+
+    wheel: Final = Path(sys.argv[1])
+    with tempfile.TemporaryDirectory() as temporary_directory, zipfile.ZipFile(wheel) as archive:
+        native_members: Final = tuple(
+            member
+            for member in archive.infolist()
+            if member.filename.startswith("litellm/rust_bridge/_native.") and member.filename.endswith(".so")
+        )
+        if len(native_members) != 1:
+            sys.stderr.write(f"expected one native extension, found {len(native_members)}\n")
+            return 1
+
+        native_path: Final = Path(temporary_directory) / Path(native_members[0].filename).name
+        native_path.write_bytes(archive.read(native_members[0]))
+        result: Final = subprocess.run((sys.executable, "-c", CHILD_SCRIPT, str(native_path)), check=False)
+
+    if result.returncode != 0:
+        sys.stderr.write(f"native wheel smoke test exited with status {result.returncode}\n")
+        return 1
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

.github/scripts/verify_linux_native_wheel.py

@@ -0,0 +1,282 @@
+from __future__ import annotations
+
+import importlib.util
+import os
+import re
+import subprocess
+import sys
+import zipfile
+from collections.abc import Callable, Mapping, Sequence
+from itertools import product
+from pathlib import Path, PurePosixPath
+from types import MappingProxyType, ModuleType
+from typing import Final, Protocol
+
+EXPECTED_PYTHON_TAG: Final = "cp310"
+EXPECTED_ABI_TAG: Final = "abi3"
+EXPECTED_PLATFORM_TAG: Final = "linux_x86_64"
+
+
+class CommandRunner(Protocol):
+    def __call__(
+        self,
+        command: tuple[str, ...],
+        *,
+        check: bool,
+        capture_output: bool,
+        text: bool,
+    ) -> subprocess.CompletedProcess[str]: ...
+
+
+def _run_command(
+    command: tuple[str, ...],
+    *,
+    check: bool,
+    capture_output: bool,
+    text: bool,
+) -> subprocess.CompletedProcess[str]:
+    return subprocess.run(command, check=check, capture_output=capture_output, text=text)
+
+
+def _dist_info_directory(member: zipfile.ZipInfo) -> str | None:
+    parts: Final = PurePosixPath(member.filename).parts
+    if not parts or not parts[0].endswith(".dist-info"):
+        return None
+    return parts[0]
+
+
+def _wheel_metadata_tags(archive: zipfile.ZipFile, members: tuple[zipfile.ZipInfo, ...]) -> tuple[str, ...]:
+    if len(members) != 1:
+        return ()
+    lines: Final = archive.read(members[0]).splitlines()
+    return tuple(line.removeprefix(b"Tag:").strip().decode("ascii") for line in lines if line.startswith(b"Tag:"))
+
+
+def _load_native_module(native_path: Path) -> ModuleType | None:
+    module_spec: Final = importlib.util.spec_from_file_location("litellm.rust_bridge._native", native_path)
+    if module_spec is None or module_spec.loader is None:
+        return None
+    try:
+        native_module: Final = importlib.util.module_from_spec(module_spec)
+        module_spec.loader.exec_module(native_module)
+    except Exception as error:  # noqa: BLE001  # native module initialization can raise arbitrary exceptions
+        sys.stderr.write(f"native module load failed: {error}\n")
+        return None
+    return native_module
+
+
+def main(
+    argv: Sequence[str] | None = None,
+    environment: Mapping[str, str] | None = None,
+    load_native_module: Callable[[Path], ModuleType | None] = _load_native_module,
+    run_command: CommandRunner = _run_command,
+) -> int:
+    arguments: Final = tuple(sys.argv if argv is None else argv)
+    resolved_environment: Final = os.environ if environment is None else environment
+    if len(arguments) != 2:
+        sys.stderr.write(f"usage: {Path(arguments[0]).name} WHEEL\n")
+        return 2
+
+    wheel: Final = Path(arguments[1])
+    wheel_tags: Final = wheel.stem.rsplit("-", maxsplit=3)
+    if len(wheel_tags) != 4:
+        sys.stderr.write(f"cannot parse wheel tags from {wheel.name}\n")
+        return 1
+
+    wheel_identity: Final = wheel_tags[0].split("-")
+    if len(wheel_identity) != 2 or wheel_identity[0] != "litellm" or not wheel_identity[1]:
+        sys.stderr.write(f"unexpected wheel identity: {wheel_tags[0]}\n")
+        return 1
+
+    expected_dist_info_directory: Final = f"{wheel_tags[0]}.dist-info"
+    expected_dist_info_directories: Final = frozenset((expected_dist_info_directory,))
+    python_tag: Final = wheel_tags[1]
+    abi_tag: Final = wheel_tags[2]
+    platform_tag: Final = wheel_tags[3]
+    expanded_filename_tags: Final = frozenset(
+        "-".join(tag) for tag in product(python_tag.split("."), abi_tag.split("."), platform_tag.split("."))
+    )
+
+    with zipfile.ZipFile(wheel) as archive:
+        wheel_members: Final = archive.infolist()
+        dist_info_directories: Final = frozenset(
+            directory for member in wheel_members if (directory := _dist_info_directory(member)) is not None
+        )
+        required_dist_info_files: Final = ("METADATA", "RECORD", "WHEEL")
+        dist_info_file_counts: Final = MappingProxyType(
+            {
+                filename: sum(
+                    member.filename == f"{expected_dist_info_directory}/{filename}" for member in wheel_members
+                )
+                for filename in required_dist_info_files
+            }
+        )
+        wheel_metadata_members: Final = tuple(
+            member for member in wheel_members if member.filename == f"{expected_dist_info_directory}/WHEEL"
+        )
+        wheel_metadata_tags: Final = _wheel_metadata_tags(archive, wheel_metadata_members)
+        native_members: Final = tuple(
+            member
+            for member in wheel_members
+            if member.filename.startswith("litellm/rust_bridge/_native.") and member.filename.endswith(".so")
+        )
+        if len(native_members) != 1:
+            sys.stderr.write(f"expected one native extension, found {len(native_members)}\n")
+            return 1
+
+        unexpected_members: Final = tuple(
+            member.filename
+            for member in wheel_members
+            if member.filename.endswith((".pdb", ".dwp", ".rlib", ".rmeta", "Cargo.toml", "Cargo.lock"))
+            or any(part.endswith(".dSYM") for part in PurePosixPath(member.filename).parts)
+        )
+        native_member: Final = native_members[0]
+        uncompressed_wheel_size: Final = sum(member.file_size for member in wheel_members)
+        native_path: Final = wheel.parent / "native" / Path(native_member.filename).name
+        native_path.parent.mkdir(parents=True, exist_ok=True)
+        native_path.write_bytes(archive.read(native_member))
+
+    wheel_metadata_tags_match: Final = (
+        len(wheel_metadata_tags) == len(expanded_filename_tags)
+        and frozenset(wheel_metadata_tags) == expanded_filename_tags
+    )
+    commit_sha: Final = resolved_environment.get(
+        "RELEASE_WHEEL_COMMIT_SHA", resolved_environment.get("GITHUB_SHA", "unknown")
+    )
+    rustc_version: Final = run_command(
+        ("rustc", "--version"),
+        check=True,
+        capture_output=True,
+        text=True,
+    ).stdout.strip()
+    pyproject: Final = (Path(__file__).parents[2] / "pyproject.toml").read_text()
+    maturin_match: Final = re.search(r'"maturin==([^";]+)', pyproject)
+    if maturin_match is None:
+        sys.stderr.write("build-system does not pin an exact Maturin version\n")
+        return 1
+
+    maturin_version: Final = maturin_match.group(1)
+    native_percentage: Final = native_member.file_size / uncompressed_wheel_size * 100
+    size_report: Final = "\n".join(
+        (
+            "## Native wheel build report",
+            "",
+            "| Build | Value |",
+            "| --- | --- |",
+            f"| Commit | `{commit_sha}` |",
+            f"| Platform | `{platform_tag}` |",
+            f"| Python ABI | `{python_tag}-{abi_tag}` |",
+            f"| Rust compiler | `{rustc_version}` |",
+            f"| Maturin | `{maturin_version}` |",
+            "| Cargo profile | `release` |",
+            "",
+            "| Artifact | Size |",
+            "| --- | ---: |",
+            f"| Compressed wheel | {wheel.stat().st_size / 1_000_000:.2f} MB |",
+            f"| Uncompressed wheel | {uncompressed_wheel_size / 1_000_000:.2f} MB |",
+            f"| Native extension | {native_member.file_size / 1_000_000:.2f} MB |",
+            f"| Native share | {native_percentage:.2f}% |",
+            "",
+        )
+    )
+    summary_path: Final = resolved_environment.get("GITHUB_STEP_SUMMARY")
+    if summary_path is None:
+        sys.stdout.write(size_report)
+    else:
+        Path(summary_path).write_text(size_report)
+
+    sections: Final = run_command(
+        ("readelf", "--sections", "--wide", str(native_path)),
+        check=True,
+        capture_output=True,
+        text=True,
+    ).stdout
+    debug_sections: Final = tuple(section for section in (".debug_", ".zdebug_") if section in sections)
+    debug_sections_absent: Final = not debug_sections
+    static_symbol_table_absent: Final = ".symtab" not in sections
+
+    dynamic_symbols: Final = run_command(
+        ("readelf", "--dyn-syms", "--wide", str(native_path)),
+        check=True,
+        capture_output=True,
+        text=True,
+    ).stdout
+    extension_entry_point_present: Final = "PyInit__native" in dynamic_symbols
+    native_module: Final = load_native_module(native_path)
+    native_module_loads: Final = native_module is not None
+    panic_test_hook_absent: Final = native_module is not None and not hasattr(native_module, "_panic_for_test")
+    native_size_limit: Final = 20_000_000
+    native_size_within_limit: Final = native_member.file_size <= native_size_limit
+    validations: Final = (
+        (f"Python tag is {EXPECTED_PYTHON_TAG}", python_tag == EXPECTED_PYTHON_TAG),
+        (f"ABI tag is {EXPECTED_ABI_TAG}", abi_tag == EXPECTED_ABI_TAG),
+        (f"Platform tag is {EXPECTED_PLATFORM_TAG}", platform_tag == EXPECTED_PLATFORM_TAG),
+        ("Wheel dist-info directory matches the filename", dist_info_directories == expected_dist_info_directories),
+        (
+            "Required dist-info files are present exactly once",
+            all(count == 1 for count in dist_info_file_counts.values()),
+        ),
+        ("Wheel metadata tags match the filename", wheel_metadata_tags_match),
+        ("Debug sections are absent", debug_sections_absent),
+        ("Static symbol table is absent", static_symbol_table_absent),
+        ("Python extension entry point is present", extension_entry_point_present),
+        ("Native module loads", native_module_loads),
+        ("Production module omits the panic test hook", panic_test_hook_absent),
+        ("Native extension does not exceed 20 MB", native_size_within_limit),
+        ("Wheel contents are valid", not unexpected_members),
+    )
+
+    verified_report: Final = size_report + "\n".join(
+        ("", "| Validation | Expected | Result |", "| --- | --- | :---: |")
+        + tuple(f"| {label} | Yes | {'O' if passed else 'X'} |" for label, passed in validations)
+        + ("",)
+    )
+    if summary_path is not None:
+        Path(summary_path).write_text(verified_report)
+
+    invalid_dist_info_files: Final = any(count != 1 for count in dist_info_file_counts.values())
+    validation_errors: Final = tuple(
+        message
+        for failed, message in (
+            (bool(debug_sections), f"{native_member.filename} contains debug sections: {', '.join(debug_sections)}"),
+            (not static_symbol_table_absent, f"{native_member.filename} contains a static symbol table"),
+            (not extension_entry_point_present, "native extension does not export PyInit__native"),
+            (
+                python_tag != EXPECTED_PYTHON_TAG,
+                f"unexpected Python tag: expected {EXPECTED_PYTHON_TAG}, found {python_tag}",
+            ),
+            (abi_tag != EXPECTED_ABI_TAG, f"unexpected ABI tag: expected {EXPECTED_ABI_TAG}, found {abi_tag}"),
+            (
+                platform_tag != EXPECTED_PLATFORM_TAG,
+                f"unexpected platform tag: expected {EXPECTED_PLATFORM_TAG}, found {platform_tag}",
+            ),
+            (
+                dist_info_directories != expected_dist_info_directories,
+                f"unexpected dist-info directories: expected {expected_dist_info_directory}, "
+                f"found {', '.join(sorted(dist_info_directories))}",
+            ),
+            (invalid_dist_info_files, f"required dist-info file counts are invalid: {dist_info_file_counts}"),
+            (
+                not invalid_dist_info_files and not wheel_metadata_tags_match,
+                f"WHEEL tags do not match filename: expected {', '.join(sorted(expanded_filename_tags))}, "
+                f"found {', '.join(sorted(wheel_metadata_tags))}",
+            ),
+            (
+                native_module is not None and not panic_test_hook_absent,
+                "production native module exposes _panic_for_test",
+            ),
+            (
+                not native_size_within_limit,
+                f"native extension exceeds 20 MB: {native_member.file_size / 1_000_000:.2f} MB",
+            ),
+            (bool(unexpected_members), f"wheel contains unexpected build artifacts: {', '.join(unexpected_members)}"),
+        )
+        if failed
+    )
+    sys.stderr.write("".join(f"{message}\n" for message in validation_errors))
+
+    return 0 if all(passed for _, passed in validations) else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())

.github/workflows/report-rust-release-wheel.yml

@@ -0,0 +1,130 @@
+name: Report LiteLLM Rust release wheel
+
+on: # zizmor: ignore[dangerous-triggers] reporter executes no PR code and consumes no PR artifacts or outputs
+  workflow_run:
+    workflows:
+      - LiteLLM Rust
+    types:
+      - completed
+
+permissions: {}
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.workflow_run.pull_requests[0].number || github.event.workflow_run.id }}
+  cancel-in-progress: false
+
+jobs:
+  report-release-wheel:
+    name: report release wheel
+    if: >-
+      github.event.workflow_run.event == 'pull_request' &&
+      github.event.workflow_run.path == '.github/workflows/test-rust.yml' &&
+      github.event.workflow_run.head_repository.full_name == github.repository &&
+      github.event.workflow_run.pull_requests[0].number != null
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    permissions:
+      issues: write # PR comments use the issues API
+      pull-requests: read # Current-head validation rejects stale workflow runs
+
+    steps:
+      - name: Link release wheel report on PR
+        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
+        env:
+          COMMENT_MARKER: "<!-- litellm-release-wheel-size -->"
+        with:
+          script: |
+            const marker = process.env.COMMENT_MARKER;
+            const workflowRun = context.payload.workflow_run;
+            const allowedConclusions = new Set([
+              "action_required",
+              "cancelled",
+              "failure",
+              "neutral",
+              "skipped",
+              "stale",
+              "startup_failure",
+              "success",
+              "timed_out",
+            ]);
+            if (
+              !allowedConclusions.has(workflowRun.conclusion) ||
+              workflowRun.event !== "pull_request" ||
+              workflowRun.path !== ".github/workflows/test-rust.yml" ||
+              workflowRun.head_repository?.full_name !==
+                `${context.repo.owner}/${context.repo.repo}` ||
+              workflowRun.pull_requests?.length !== 1
+            ) {
+              throw new Error("unexpected source workflow");
+            }
+            const pullRequest = workflowRun.pull_requests[0];
+            const pullRequestNumber = pullRequest.number;
+            const headSha = workflowRun.head_sha;
+            const runId = workflowRun.id;
+            if (
+              !Number.isSafeInteger(pullRequestNumber) ||
+              pullRequestNumber <= 0 ||
+              !Number.isSafeInteger(runId) ||
+              runId <= 0 ||
+              !/^[0-9a-f]{40}$/.test(headSha) ||
+              pullRequest.head?.sha !== headSha
+            ) {
+              throw new Error("invalid source workflow metadata");
+            }
+            const runUrl =
+              `${context.serverUrl}/${context.repo.owner}/${context.repo.repo}` +
+              `/actions/runs/${runId}`;
+            const result =
+              workflowRun.conclusion === "success"
+                ? "successfully"
+                : `with \`${workflowRun.conclusion}\``;
+            const body = [
+              marker,
+              "## LiteLLM Rust workflow",
+              "",
+              `Workflow completed ${result} for \`${headSha}\``,
+              "",
+              `[View workflow run](${runUrl})`,
+            ].join("\n");
+            const comments = await github.paginate(github.rest.issues.listComments, {
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: pullRequestNumber,
+              per_page: 100,
+            });
+            const existing = comments.find(
+              (comment) =>
+                comment.user?.login === "github-actions[bot]" &&
+                comment.body?.startsWith(marker),
+            );
+            const currentPullRequest = (
+              await github.rest.pulls.get({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                pull_number: pullRequestNumber,
+              })
+            ).data;
+            if (
+              currentPullRequest.state !== "open" ||
+              currentPullRequest.head.repo?.full_name !==
+                `${context.repo.owner}/${context.repo.repo}` ||
+              currentPullRequest.head.sha !== headSha
+            ) {
+              core.info("source workflow no longer matches the current pull request head");
+              return;
+            }
+            if (existing) {
+              await github.rest.issues.updateComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                comment_id: existing.id,
+                body,
+              });
+            } else {
+              await github.rest.issues.createComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                issue_number: pullRequestNumber,
+                body,
+              });
+            }

.github/workflows/test-rust.yml

@@ -4,6 +4,11 @@ on:
   push:
     paths:
       - "litellm-rust/**"
+      - ".cargo/**"
+      - "pyproject.toml"
+      - "rust-toolchain.toml"
+      - ".github/scripts/smoke_test_native_wheel.py"
+      - ".github/scripts/verify_linux_native_wheel.py"
       - ".github/workflows/test-rust.yml"
   pull_request:
     branches:
@@ -13,6 +18,11 @@ on:
       - "litellm_**"
     paths:
       - "litellm-rust/**"
+      - ".cargo/**"
+      - "pyproject.toml"
+      - "rust-toolchain.toml"
+      - ".github/scripts/smoke_test_native_wheel.py"
+      - ".github/scripts/verify_linux_native_wheel.py"
       - ".github/workflows/test-rust.yml"
 
 permissions:
@@ -40,9 +50,7 @@ jobs:
           persist-credentials: false
 
       - name: Set up Rust
-        run: |
-          rustup toolchain install stable --profile minimal --component clippy,rustfmt
-          rustup default stable
+        run: rustup toolchain install
 
       - name: Cache Cargo registry and target
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
@@ -51,7 +59,7 @@ jobs:
             ~/.cargo/registry
             ~/.cargo/git
             litellm-rust/target
-          key: ${{ runner.os }}-cargo-${{ hashFiles('litellm-rust/Cargo.lock') }}
+          key: ${{ runner.os }}-cargo-${{ hashFiles('rust-toolchain.toml', 'litellm-rust/Cargo.lock') }}
           restore-keys: |
             ${{ runner.os }}-cargo-
 
@@ -69,3 +77,47 @@ jobs:
 
       - name: Run core tests with Bedrock auth
         run: cargo test -p litellm-core --features bedrock-auth --locked
+
+  release-wheel:
+    name: release wheel
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    permissions:
+      contents: read
+    env:
+      CARGO_TERM_COLOR: always
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Set up uv
+        uses: ./.github/actions/setup-uv-with-retries
+        with:
+          version: "0.10.9"
+
+      - name: Set up Rust
+        run: rustup toolchain install
+
+      - name: Build release wheel
+        run: uv build --wheel --out-dir dist
+
+      - name: Build panic contract wheel
+        run: >-
+          uv build --wheel --out-dir panic-dist
+          --config-setting "maturin.build-args=--features panic-test,extension-module"
+
+      - name: Smoke-test native panic unwinding
+        run: python .github/scripts/smoke_test_native_wheel.py panic-dist/*.whl
+
+      - name: Verify stripped native extension
+        env:
+          RELEASE_WHEEL_COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+        run: python .github/scripts/verify_linux_native_wheel.py dist/*.whl

basedpyright-code-budget.json

@@ -1,6 +1,6 @@
 {
   "reportAny": {
-    "limit": 14076
+    "limit": 14074
   },
   "reportArgumentType": {
     "limit": 2216
@@ -24,7 +24,7 @@
     "limit": 19
   },
   "reportExplicitAny": {
-    "limit": 4128
+    "limit": 4125
   },
   "reportFunctionMemberAccess": {
     "limit": 7
@@ -108,10 +108,10 @@
     "limit": 38350
   },
   "reportUnknownParameterType": {
-    "limit": 19626
+    "limit": 19625
   },
   "reportUnknownVariableType": {
-    "limit": 29890
+    "limit": 29877
   },
   "reportUnnecessaryCast": {
     "limit": 111
@@ -138,7 +138,7 @@
     "limit": 138
   },
   "reportUnusedImport": {
-    "limit": 543
+    "limit": 542
   },
   "reportUnusedVariable": {
     "limit": 137

enterprise/litellm_enterprise/proxy/management_endpoints/internal_user_endpoints.py

@@ -39,9 +39,9 @@ async def available_enterprise_users(
 
     if not premium_user:
         # check if SSO is enabled - show 5 user limit
-        from litellm.proxy.auth.auth_utils import _has_user_setup_sso
+        from litellm.proxy.auth.auth_utils import has_user_setup_sso
 
-        if _has_user_setup_sso():
+        if has_user_setup_sso():
             premium_user_data = EnterpriseLicenseData(
                 max_users=5,
             )

litellm-rust/AGENTS.md

@@ -1,6 +1,6 @@
 # AGENTS.md
 
-litellm-rust has exactly THREE crates. A crate is a LAYER, not a route. Routes (ocr, realtime, chat) and providers (mistral, openai) are MODULES inside the layers.
+litellm-rust has four crates. A crate is a layer or shared foundation, not a route. Routes (ocr, realtime, chat) and providers (mistral, openai) are modules inside the layers.
 
 ## Crates
 
@@ -8,9 +8,10 @@ litellm-rust has exactly THREE crates. A crate is a LAYER, not a route. Routes (
 |-------|------|
 | litellm-core | The LiteLLM SDK in Rust. One public entrypoint per top-level call (`messages::messages()`), owning types, transforms, provider resolution, auth, and the provider HTTP call. Call it, get a typed response. |
 | litellm-ai-gateway | The axum server (behind the `server` feature) plus the WebSocket hosts. Translates HTTP/WS to core entrypoints; owns no provider logic and no handlers. |
-| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
+| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
+| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. Owns API registration, domain wiring, and Python exception mapping. |
 
-Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
+Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
 
 ## Where a route lives
 
@@ -28,7 +29,7 @@ core/src/messages/
 
 Handlers never live in `ai-gateway`. `ocr`, `audio_transcription`, and `realtime` are still hosted there from before this rule; they move to `core` as they are touched.
 
-Adding a crate: default to a MODULE. New crate ONLY on a real trigger — separate artifact (binary/cdylib), proc-macro, shared foundation, or publishable standalone. A new provider or route is none of these.
+Adding a crate: default to a module. A new crate requires a real trigger: separate artifact (binary/cdylib), proc-macro, shared foundation, or publishable standalone. A new provider or route is none of these.
 
 Adding a crate fails crates/core/tests/workspace_crate_allowlist.rs until you update its allowlist and this file — intentional.
 

litellm-rust/CLAUDE.md

@@ -21,12 +21,13 @@ variants of it. The test for a good abstraction is that adding the next provider
 is a few declarative lines, not a new file of duplicated flow. Only diverge from
 the base when behavior is genuinely different, and say so explicitly in the PR.
 
-## Crates (exactly three — see AGENTS.md)
+## Crates (see AGENTS.md)
 
 `litellm-core` **is** the LiteLLM SDK in Rust: it makes the LLM call.
 `litellm-ai-gateway` is an HTTP/WebSocket server in front of it, and
-`litellm-python-bridge` exposes it to the Python SDK. A crate is a **layer**, not
-a route — add modules, not crates.
+`litellm-python-bridge` exposes it to the Python SDK. `litellm-python-interop`
+holds domain-neutral PyO3 primitives shared by Python-facing Rust code. A crate
+is a layer or shared foundation, not a route; add modules, not crates.
 
 ## Core Boundary
 
@@ -175,7 +176,7 @@ cd litellm-rust
 cargo fmt --check
 # the ai-gateway binary + server code is behind the `server` feature
 cargo clippy -p litellm-ai-gateway --all-targets --features server -- -D warnings
-cargo clippy -p litellm-core -p litellm-python-bridge --all-targets -- -D warnings
+cargo clippy -p litellm-core -p litellm-python-interop -p litellm-python-bridge --all-targets -- -D warnings
 cargo test --workspace
 ```
 

litellm-rust/Cargo.lock

@@ -919,6 +919,12 @@ version = "0.3.33"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b231ed28831efb4a61a08580c4bc233ec56bc009f4cd8f52da2c3cb97df0c109"
 
+[[package]]
+name = "futures-timer"
+version = "3.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "af43fadb8a98512d547e37b4e92e0ced13e205c061b87b4623eff01d918d6968"
+
 [[package]]
 name = "futures-util"
 version = "0.3.33"
@@ -972,6 +978,12 @@ dependencies = [
  "wasm-bindgen",
 ]
 
+[[package]]
+name = "glob"
+version = "0.3.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e4eba85ea1d0a966a983acd07deee566e67395d2d96b6fb39e62b5a833f1eb0b"
+
 [[package]]
 name = "h2"
 version = "0.3.27"
@@ -1432,14 +1444,24 @@ dependencies = [
  "criterion",
  "litellm-ai-gateway",
  "litellm-core",
+ "litellm-python-interop",
  "pyo3",
  "pyo3-async-runtimes",
- "pythonize",
- "serde",
  "serde_json",
  "tokio",
 ]
 
+[[package]]
+name = "litellm-python-interop"
+version = "0.1.0"
+dependencies = [
+ "pyo3",
+ "pythonize",
+ "rstest",
+ "serde",
+ "serde_json",
+]
+
 [[package]]
 name = "litemap"
 version = "0.8.2"
@@ -1627,6 +1649,15 @@ dependencies = [
  "zerocopy",
 ]
 
+[[package]]
+name = "proc-macro-crate"
+version = "3.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e67ba7e9b2b56446f1d419b1d807906278ffa1a658a8a5d8a39dcb1f5a78614f"
+dependencies = [
+ "toml_edit",
+]
+
 [[package]]
 name = "proc-macro2"
 version = "1.0.107"
@@ -1899,6 +1930,12 @@ version = "0.8.11"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d6f6ff9a378485b298a5286656da665ba74413d36db0979633275d2e708145d4"
 
+[[package]]
+name = "relative-path"
+version = "1.9.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba39f3699c378cd8970968dcbff9c43159ea4cfbd88d43c00b22f2ef10a435d2"
+
 [[package]]
 name = "reqwest"
 version = "0.12.28"
@@ -1956,6 +1993,35 @@ dependencies = [
  "windows-sys 0.52.0",
 ]
 
+[[package]]
+name = "rstest"
+version = "0.26.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f5a3193c063baaa2a95a33f03035c8a72b83d97a54916055ba22d35ed3839d49"
+dependencies = [
+ "futures-timer",
+ "futures-util",
+ "rstest_macros",
+]
+
+[[package]]
+name = "rstest_macros"
+version = "0.26.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9c845311f0ff7951c5506121a9ad75aec44d083c31583b2ea5a30bcb0b0abba0"
+dependencies = [
+ "cfg-if",
+ "glob",
+ "proc-macro-crate",
+ "proc-macro2",
+ "quote",
+ "regex",
+ "relative-path",
+ "rustc_version",
+ "syn 2.0.119",
+ "unicode-ident",
+]
+
 [[package]]
 name = "rustc-hash"
 version = "2.1.3"
@@ -2488,6 +2554,36 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "toml_datetime"
+version = "1.1.1+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3165f65f62e28e0115a00b2ebdd37eb6f3b641855f9d636d3cd4103767159ad7"
+dependencies = [
+ "serde_core",
+]
+
+[[package]]
+name = "toml_edit"
+version = "0.25.13+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6975367e4d2ef766d86af01ffad14b622fecc8d4357a998fbc4deb6e9bacaf9b"
+dependencies = [
+ "indexmap",
+ "toml_datetime",
+ "toml_parser",
+ "winnow",
+]
+
+[[package]]
+name = "toml_parser"
+version = "1.1.3+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1d38ac1cf9b95face32296c0a3ede1fdc270627c9d9c02a7274dd6d960dc4d56"
+dependencies = [
+ "winnow",
+]
+
 [[package]]
 name = "tower"
 version = "0.5.3"
@@ -2903,6 +2999,15 @@ version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec"
 
+[[package]]
+name = "winnow"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "23b97319f7b8343df12cc98938e5c3eb436064524c8d2b4e30a1d3a36eecdf81"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "writeable"
 version = "0.6.3"

litellm-rust/Cargo.toml

@@ -2,6 +2,7 @@
 members = [
     "crates/core",
     "crates/ai-gateway",
+    "crates/python-interop",
     "crates/python-bridge",
 ]
 resolver = "2"
@@ -15,12 +16,14 @@ repository = "https://github.com/BerriAI/litellm"
 [workspace.dependencies]
 litellm-core = { path = "crates/core" }
 litellm-ai-gateway = { path = "crates/ai-gateway", default-features = false }
+litellm-python-interop = { path = "crates/python-interop" }
 axum = "0.7"
 pyo3 = "0.29.0"
 pyo3-async-runtimes = { version = "0.29.0", features = ["tokio-runtime"] }
 pythonize = "0.29.0"
 rand = "0.8"
 reqwest = { version = "0.12", default-features = false, features = ["blocking", "json", "rustls-tls", "http2", "stream"] }
+rstest = "0.26.1"
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 sha2 = "0.10"

litellm-rust/README.md

@@ -26,9 +26,10 @@ coverage and production evidence.
 |-------|------|
 | litellm-core | The SDK. Per-route entrypoints (`messages::messages()`), types, provider transforms (modules under `providers/`), provider resolution, auth, the provider HTTP call, and the router. |
 | litellm-ai-gateway | The axum server (behind the `server` feature) and WebSocket hosts. Translates HTTP/WS to core entrypoints; no provider handlers. |
-| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
+| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
+| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. Owns API registration, domain wiring, and Python exception mapping. |
 
-Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
+Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
 
 ## Layout
 
@@ -38,7 +39,8 @@ crates/
     src/messages/   mod.rs (entrypoint), types, transformation, prepare, handler, client
     src/providers/anthropic/messages/transformation.rs
   ai-gateway/     Axum server + WebSocket hosts; calls core entrypoints.
-  python-bridge/  PyO3 bridge for Python LiteLLM.
+  python-interop/ Domain-neutral PyO3 conversion and GIL primitives.
+  python-bridge/  PyO3 API adapter for Python LiteLLM.
 ```
 
 The folder shape follows the Python provider tree:

litellm-rust/crates/CODING_STANDARDS/PROVIDER_CODING_STANDARDS.md

@@ -54,6 +54,6 @@ Rules for adding or changing an LLM provider/route in `litellm-rust`. `messages`
     cd litellm-rust
     cargo fmt --check
     cargo clippy -p litellm-ai-gateway --all-targets --features server -- -D warnings
-    cargo clippy -p litellm-core -p litellm-python-bridge --all-targets -- -D warnings
+    cargo clippy -p litellm-core -p litellm-python-interop -p litellm-python-bridge --all-targets -- -D warnings
     cargo test --workspace
     ```

litellm-rust/crates/ai-gateway/README.md

@@ -6,15 +6,16 @@ dials OpenAI upstream, and splices the two sockets frame-by-frame.
 
 ## Crates
 
-`litellm-rust` is exactly three crates (a crate is a **layer**, not a route):
+`litellm-rust` has four crates. A crate is a layer or shared foundation, not a route:
 
 | Crate | Role |
 |-------|------|
 | litellm-core | The LiteLLM SDK in Rust — per-route entrypoints (`messages::messages()`) that resolve the provider, transform, and make the call; plus types, provider transforms, and the router. |
 | litellm-ai-gateway | The Axum server (behind the `server` feature) and WebSocket hosts. Translates HTTP/WS to core entrypoints; no provider handlers. |
-| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
+| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
+| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. |
 
-Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
+Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
 
 - **Client endpoint:** `wss://<host>/v1/realtime?model=<model>` (WebSocket)
 - **Auth:** `Authorization: Bearer $LITELLM_MASTER_KEY` (fails closed if unset)

litellm-rust/crates/ai-gateway/src/audio_transcription/common_utils.rs

@@ -1,10 +1,8 @@
-use std::collections::BTreeMap;
-
-use litellm_core::CoreResult;
 use litellm_core::audio_transcription::transformation::AudioTranscriptionProviderConfig;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::providers::bedrock::audio_transcription::BEDROCK_AUDIO_TRANSCRIPTION_CONFIG;
 use serde_json::{Map, Value};
+use std::collections::BTreeMap;
 
 pub(super) fn audio_transcription_provider_config(
     provider: &str,
@@ -17,7 +15,7 @@ pub(super) fn audio_transcription_provider_config(
 
 pub(super) fn string_headers(
     headers: Option<Map<String, Value>>,
-) -> CoreResult<BTreeMap<String, String>> {
+) -> Result<BTreeMap<String, String>, Error> {
     headers
         .unwrap_or_default()
         .into_iter()
@@ -26,7 +24,7 @@ pub(super) fn string_headers(
                 .as_str()
                 .map(|value| (key.clone(), value.to_string()))
                 .ok_or_else(|| {
-                    CoreError::InvalidRequest(format!(
+                    Error::InvalidRequest(format!(
                         "audio transcription extra_headers.{key} must be a string"
                     ))
                 })

litellm-rust/crates/ai-gateway/src/audio_transcription/handler.rs

@@ -1,11 +1,9 @@
-use std::time::SystemTime;
-
-use litellm_core::CoreResult;
 use litellm_core::audio_transcription::transformation::AudioTranscriptionAuth;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::providers::bedrock::audio_transcription::aws_auth_config;
 use litellm_core::providers::bedrock::aws_base::{resolve_credentials, sign_bedrock_post};
 use serde_json::Value;
+use std::time::SystemTime;
 
 use super::common_utils::truncate_error_body;
 use super::types::ProviderAudioTranscriptionRequest;
@@ -13,10 +11,9 @@ use crate::client::http_client;
 
 pub(crate) async fn execute_audio_transcription_provider_call(
     request: ProviderAudioTranscriptionRequest,
-) -> CoreResult<Value> {
-    let body = serde_json::to_vec(&request.body).map_err(|error| {
-        CoreError::InvalidRequest(format!("invalid audio request body: {error}"))
-    })?;
+) -> Result<Value, Error> {
+    let body = serde_json::to_vec(&request.body)
+        .map_err(|error| Error::InvalidRequest(format!("invalid audio request body: {error}")))?;
     let mut request_builder = http_client().post(&request.url).body(body.clone());
     for (key, value) in &request.upstream_headers {
         request_builder = request_builder.header(key, value);
@@ -27,21 +24,20 @@ pub(crate) async fn execute_audio_transcription_provider_call(
     let response = request_builder
         .send()
         .await
-        .map_err(|error| CoreError::Network(error.to_string()))?;
+        .map_err(|error| Error::Network(error.to_string()))?;
     let status = response.status();
     let text = response
         .text()
         .await
-        .map_err(|error| CoreError::Network(error.to_string()))?;
+        .map_err(|error| Error::Network(error.to_string()))?;
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
-    let response_json: Value = serde_json::from_str(&text).map_err(|error| {
-        CoreError::InvalidResponse(format!("invalid audio response JSON: {error}"))
-    })?;
+    let response_json: Value = serde_json::from_str(&text)
+        .map_err(|error| Error::InvalidResponse(format!("invalid audio response JSON: {error}")))?;
     Ok(request
         .config
         .transform_transcription_response(&request.model, response_json)?
@@ -51,14 +47,13 @@ pub(crate) async fn execute_audio_transcription_provider_call(
 pub(crate) async fn sign_request(
     request: &ProviderAudioTranscriptionRequest,
     optional_params: &serde_json::Map<String, Value>,
-) -> CoreResult<ProviderAudioTranscriptionRequest> {
+) -> Result<ProviderAudioTranscriptionRequest, Error> {
     let env_lookup = environment_lookup;
     let auth = request
         .config
         .auth_strategy(&request.model, optional_params, &env_lookup)?;
-    let body = serde_json::to_vec(&request.body).map_err(|error| {
-        CoreError::InvalidRequest(format!("invalid audio request body: {error}"))
-    })?;
+    let body = serde_json::to_vec(&request.body)
+        .map_err(|error| Error::InvalidRequest(format!("invalid audio request body: {error}")))?;
     let mut headers = super::common_utils::string_headers(None)?;
     headers.insert("Content-Type".to_string(), "application/json".to_string());
     headers.extend(request.upstream_headers.iter().cloned());

litellm-rust/crates/ai-gateway/src/audio_transcription/hooks.rs

@@ -1,11 +1,9 @@
-use std::future::Future;
-use std::pin::Pin;
-
-use litellm_core::CoreResult;
 use litellm_core::audio_transcription::transformation::AudioTranscriptionAuth;
 use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use serde_json::{Map, Value, json};
+use std::future::Future;
+use std::pin::Pin;
 
 use super::common_utils::{audio_transcription_provider_config, has_header, string_headers};
 use super::handler::sign_request;
@@ -26,7 +24,7 @@ pub(crate) struct AudioTranscriptionLifecycleHooks {
     request_metadata: RequestMetadata,
 }
 
-type AudioFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
+type AudioFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
 type AudioLogFuture<'a> = Pin<Box<dyn Future<Output = ()> + Send + 'a>>;
 
 impl AudioTranscriptionLifecycleHooks {
@@ -45,7 +43,7 @@ impl AudioTranscriptionLifecycleHooks {
     async fn run_pre_call_guardrails(
         &self,
         request: PreparedAudioTranscriptionRequest,
-    ) -> CoreResult<PreparedAudioTranscriptionRequest> {
+    ) -> Result<PreparedAudioTranscriptionRequest, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(request);
         }
@@ -63,17 +61,17 @@ impl AudioTranscriptionLifecycleHooks {
             .await
             .map_err(guardrail_error_to_core_error)?;
         let Value::Object(mut data) = guardrail_request.data else {
-            return Err(CoreError::InvalidRequest(
+            return Err(Error::InvalidRequest(
                 "audio transcription pre_call guardrail must return an object".to_string(),
             ));
         };
         let audio = data.remove("audio").ok_or_else(|| {
-            CoreError::InvalidRequest("audio transcription guardrail removed audio".to_string())
+            Error::InvalidRequest("audio transcription guardrail removed audio".to_string())
         })?;
         let optional_params = match data.remove("optional_params") {
             Some(Value::Object(value)) => value,
             Some(_) => {
-                return Err(CoreError::InvalidRequest(
+                return Err(Error::InvalidRequest(
                     "audio transcription optional_params must be an object".to_string(),
                 ));
             }
@@ -89,9 +87,9 @@ impl AudioTranscriptionLifecycleHooks {
     async fn prepare_provider_request(
         &self,
         request: PreparedAudioTranscriptionRequest,
-    ) -> CoreResult<ProviderAudioTranscriptionRequest> {
+    ) -> Result<ProviderAudioTranscriptionRequest, Error> {
         let config = audio_transcription_provider_config(&request.custom_llm_provider)
-            .ok_or_else(|| CoreError::InvalidProvider(request.custom_llm_provider.clone()))?;
+            .ok_or_else(|| Error::InvalidProvider(request.custom_llm_provider.clone()))?;
         let env_lookup = super::handler::environment_lookup;
         let headers = string_headers(request.extra_headers)?;
         let url = config.complete_url(
@@ -135,7 +133,7 @@ impl AudioTranscriptionLifecycleHooks {
     async fn run_during_call_guardrails(
         &self,
         request: ProviderAudioTranscriptionRequest,
-    ) -> CoreResult<ProviderAudioTranscriptionRequest> {
+    ) -> Result<ProviderAudioTranscriptionRequest, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(request);
         }
@@ -153,12 +151,12 @@ impl AudioTranscriptionLifecycleHooks {
             .await
             .map_err(guardrail_error_to_core_error)?;
         let Value::Object(mut data) = guardrail_request.data else {
-            return Err(CoreError::InvalidRequest(
+            return Err(Error::InvalidRequest(
                 "audio transcription during_call guardrail must return an object".to_string(),
             ));
         };
         let body = data.remove("body").ok_or_else(|| {
-            CoreError::InvalidRequest("audio transcription guardrail removed body".to_string())
+            Error::InvalidRequest("audio transcription guardrail removed body".to_string())
         })?;
         Ok(ProviderAudioTranscriptionRequest { body, ..request })
     }
@@ -241,7 +239,7 @@ impl CallLifecycleHooks<PreparedAudioTranscriptionRequest, ProviderAudioTranscri
     fn async_log_failure_event<'a>(
         &'a self,
         context: &'a CallLifecycleContext,
-        error: &'a CoreError,
+        error: &'a Error,
         timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a> {
         Box::pin(async move {
@@ -281,22 +279,22 @@ fn guardrail_context(metadata: &RequestMetadata) -> GuardrailContext {
     }
 }
 
-fn guardrail_error_to_core_error(error: GuardrailError) -> CoreError {
-    CoreError::InvalidRequest(format!("{}: {}", error.kind, error.message))
+fn guardrail_error_to_core_error(error: GuardrailError) -> Error {
+    Error::InvalidRequest(format!("{}: {}", error.kind, error.message))
 }
 
-fn core_error_kind(error: &CoreError) -> &'static str {
+fn core_error_kind(error: &Error) -> &'static str {
     match error {
-        CoreError::Auth(_) => "AuthError",
-        CoreError::InvalidProvider(_) => "InvalidProvider",
-        CoreError::InvalidRequest(_) => "InvalidRequest",
-        CoreError::InvalidType { .. } => "InvalidType",
-        CoreError::MissingField(_) => "MissingField",
-        CoreError::Http { .. } => "HttpError",
-        CoreError::InvalidResponse(_) => "InvalidResponse",
-        CoreError::Network(_) => "NetworkError",
-        CoreError::Connect(_) => "ConnectError",
-        CoreError::Routing(_) => "RoutingError",
-        CoreError::Unsupported(_) => "UnsupportedRequest",
+        Error::Auth(_) => "AuthError",
+        Error::InvalidProvider(_) => "InvalidProvider",
+        Error::InvalidRequest(_) => "InvalidRequest",
+        Error::InvalidType { .. } => "InvalidType",
+        Error::MissingField(_) => "MissingField",
+        Error::Http { .. } => "HttpError",
+        Error::InvalidResponse(_) => "InvalidResponse",
+        Error::Network(_) => "NetworkError",
+        Error::Connect(_) => "ConnectError",
+        Error::Routing(_) => "RoutingError",
+        Error::Unsupported(_) => "UnsupportedRequest",
     }
 }

litellm-rust/crates/ai-gateway/src/audio_transcription/mod.rs

@@ -1,4 +1,4 @@
-use litellm_core::CoreResult;
+use litellm_core::Error;
 use litellm_core::call_lifecycle::CallLifecycle;
 use serde_json::Value;
 
@@ -13,7 +13,7 @@ pub use types::AudioTranscriptionRequest;
 use handler::execute_audio_transcription_provider_call;
 use prepare::{PreparedAudioTranscriptionCall, prepare_audio_transcription_call};
 
-pub async fn audio_transcription(request: AudioTranscriptionRequest<'_>) -> CoreResult<Value> {
+pub async fn audio_transcription(request: AudioTranscriptionRequest<'_>) -> Result<Value, Error> {
     let PreparedAudioTranscriptionCall { request, hooks } =
         prepare_audio_transcription_call(request);
     CallLifecycle::default()

litellm-rust/crates/ai-gateway/src/io/realtime.rs

@@ -15,8 +15,7 @@ use std::time::Duration;
 
 use futures_util::stream::{SplitSink, SplitStream};
 use futures_util::{Sink, SinkExt, Stream, StreamExt};
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::realtime::transformation::RealtimeProviderConfig;
 use litellm_core::realtime::types::RealtimeEvent;
 use tokio::net::TcpStream;
@@ -48,7 +47,7 @@ pub(crate) type UpstreamRx = SplitStream<UpstreamWs>;
 /// Resolve the OpenAI API key from the explicit param or the environment.
 ///
 /// Blank/whitespace values are treated as absent (guard at resolution time).
-pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
+pub(crate) fn resolve_api_key(api_key: Option<&str>) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|key| !key.is_empty())
@@ -58,7 +57,7 @@ pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
                 .ok()
                 .filter(|key| !key.trim().is_empty())
         })
-        .ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
+        .ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
 }
 
 /// Open the upstream WebSocket to OpenAI for `(model, api_key, api_base)`.
@@ -70,24 +69,24 @@ pub(crate) async fn dial_upstream(
     model: &str,
     api_key: &str,
     api_base: Option<&str>,
-) -> CoreResult<UpstreamWs> {
+) -> Result<UpstreamWs, Error> {
     let url = OPENAI_REALTIME_CONFIG.complete_url(api_base, model);
 
     let mut request = url
         .as_str()
         .into_client_request()
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     // GA realtime: only Authorization. The legacy OpenAI-Beta header triggers
     // beta_api_shape_disabled, so we do not send it.
     request.headers_mut().insert(
         AUTHORIZATION,
         HeaderValue::from_str(&format!("Bearer {api_key}"))
-            .map_err(|err| CoreError::Auth(err.to_string()))?,
+            .map_err(|err| Error::Auth(err.to_string()))?,
     );
 
     let (upstream, _response) = connect_async(request)
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     Ok(upstream)
 }
 
@@ -96,22 +95,22 @@ pub(crate) async fn dial_upstream(
 /// Used by the pool to pre-read OpenAI's unprompted `session.created`. Returns an
 /// error on a non-text frame, a closed socket, or undecodable JSON so the pool can
 /// discard a misbehaving socket rather than warm it.
-pub(crate) async fn read_event(upstream_rx: &mut UpstreamRx) -> CoreResult<RealtimeEvent> {
+pub(crate) async fn read_event(upstream_rx: &mut UpstreamRx) -> Result<RealtimeEvent, Error> {
     loop {
         let message = upstream_rx
             .next()
             .await
-            .ok_or_else(|| CoreError::Network("upstream closed before first event".to_string()))?
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .ok_or_else(|| Error::Network("upstream closed before first event".to_string()))?
+            .map_err(|err| Error::Network(err.to_string()))?;
         match message {
             Message::Text(text) => {
                 return serde_json::from_str(&text)
-                    .map_err(|err| CoreError::InvalidResponse(err.to_string()));
+                    .map_err(|err| Error::InvalidResponse(err.to_string()));
             }
             // Ignore protocol frames (ping/pong) while waiting for the first event.
             Message::Ping(_) | Message::Pong(_) => continue,
             Message::Close(_) => {
-                return Err(CoreError::Network(
+                return Err(Error::Network(
                     "upstream closed before first event".to_string(),
                 ));
             }
@@ -139,7 +138,7 @@ pub(crate) async fn splice<In, Out>(
     mut observe: impl FnMut(&RealtimeEvent) + Send,
     mut client_in: In,
     mut client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,
@@ -154,7 +153,7 @@ where
             client_out
                 .send(outbound)
                 .await
-                .map_err(|err| CoreError::Network(err.to_string()))?;
+                .map_err(|err| Error::Network(err.to_string()))?;
         }
     }
 
@@ -175,26 +174,26 @@ where
                 // inflate its own spend log. Logging observes upstream events only.
                 for outbound in config.transform_realtime_request(&event, model)?.events {
                     let payload = serde_json::to_string(&outbound)
-                        .map_err(|err| CoreError::InvalidResponse(err.to_string()))?;
+                        .map_err(|err| Error::InvalidResponse(err.to_string()))?;
                     upstream_tx
                         .send(Message::Text(payload))
                         .await
-                        .map_err(|err| CoreError::Network(err.to_string()))?;
+                        .map_err(|err| Error::Network(err.to_string()))?;
                 }
             }
             // upstream -> client
             upstream_message = upstream_rx.next() => {
                 let Some(message) = upstream_message else { break }; // upstream closed
-                match message.map_err(|err| CoreError::Network(err.to_string()))? {
+                match message.map_err(|err| Error::Network(err.to_string()))? {
                     Message::Text(text) => {
                         let event: RealtimeEvent = serde_json::from_str(&text)
-                            .map_err(|err| CoreError::InvalidResponse(err.to_string()))?;
+                            .map_err(|err| Error::InvalidResponse(err.to_string()))?;
                         observe(&event);
                         for outbound in config.transform_realtime_response(&event, model)?.events {
                             client_out
                                 .send(outbound)
                                 .await
-                                .map_err(|err| CoreError::Network(err.to_string()))?;
+                                .map_err(|err| Error::Network(err.to_string()))?;
                         }
                     }
                     Message::Close(_) => break,
@@ -225,7 +224,7 @@ pub async fn realtime<In, Out>(
     observe: impl FnMut(&RealtimeEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,
@@ -258,7 +257,7 @@ pub async fn realtime_warm<In, Out>(
     observe: impl FnMut(&RealtimeEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,

litellm-rust/crates/ai-gateway/src/io/realtime_pool.rs

@@ -28,7 +28,7 @@ use std::sync::{Arc, Mutex};
 use std::time::{Duration, Instant};
 
 use futures_util::StreamExt;
-use litellm_core::CoreResult;
+use litellm_core::Error;
 use litellm_core::realtime::types::RealtimeEvent;
 
 use crate::io::realtime::{
@@ -438,7 +438,7 @@ impl RealtimePool {
 ///
 /// `key.api_key` is already resolved (non-blank). The first frame OpenAI sends
 /// unprompted is `session.created`; we buffer exactly that and read nothing more.
-async fn warm_one(key: &UpstreamKey) -> CoreResult<WarmConnection> {
+async fn warm_one(key: &UpstreamKey) -> Result<WarmConnection, Error> {
     let upstream: UpstreamWs =
         dial_upstream(&key.model, &key.api_key, key.api_base.as_deref()).await?;
     let (tx, mut rx) = upstream.split();

litellm-rust/crates/ai-gateway/src/io/responses_ws.rs

@@ -4,10 +4,10 @@ use std::time::Duration;
 
 use futures_util::stream::{SplitSink, SplitStream};
 use futures_util::{Sink, SinkExt, Stream, StreamExt};
+use litellm_core::Error;
 use litellm_core::providers::openai::responses::transformation::OPENAI_RESPONSES_WS_CONFIG;
 use litellm_core::responses::types::ResponsesWsEvent;
 use litellm_core::responses::websocket::ResponsesWebSocketProviderConfig;
-use litellm_core::{CoreError, CoreResult};
 use tokio::net::TcpStream;
 use tokio::sync::Mutex;
 use tokio_tungstenite::tungstenite::Message;
@@ -37,51 +37,49 @@ impl ResponsesWebSocketConnection {
         url: &str,
         headers: &HashMap<String, String>,
         timeout: Option<Duration>,
-    ) -> CoreResult<Self> {
+    ) -> Result<Self, Error> {
         let mut request = url
             .into_client_request()
-            .map_err(|error| CoreError::Network(error.to_string()))?;
+            .map_err(|error| Error::Network(error.to_string()))?;
         for (name, value) in headers {
             let header_name = name
                 .parse::<HeaderName>()
-                .map_err(|error| CoreError::InvalidRequest(error.to_string()))?;
+                .map_err(|error| Error::InvalidRequest(error.to_string()))?;
             let header_value = HeaderValue::from_str(value)
-                .map_err(|error| CoreError::InvalidRequest(error.to_string()))?;
+                .map_err(|error| Error::InvalidRequest(error.to_string()))?;
             request.headers_mut().insert(header_name, header_value);
         }
         let connect = connect_async(request);
         let result = match timeout {
             Some(timeout) => tokio::time::timeout(timeout, connect).await.map_err(|_| {
-                CoreError::Network("Responses WebSocket connection timed out".to_string())
+                Error::Network("Responses WebSocket connection timed out".to_string())
             })?,
             None => connect.await,
         };
         let (socket, _) = result.map_err(|error| match error {
-            tokio_tungstenite::tungstenite::Error::Http(response) => CoreError::Http {
+            tokio_tungstenite::tungstenite::Error::Http(response) => Error::Http {
                 status: response.status().as_u16(),
                 body: String::new(),
             },
-            other => CoreError::Network(other.to_string()),
+            other => Error::Network(other.to_string()),
         })?;
         Ok(Self {
             socket: Arc::new(Mutex::new(Some(socket))),
         })
     }
 
-    pub async fn send_text(&self, text: String) -> CoreResult<()> {
+    pub async fn send_text(&self, text: String) -> Result<(), Error> {
         let mut socket = self.socket.lock().await;
         let Some(socket) = socket.as_mut() else {
-            return Err(CoreError::Network(
-                "Responses WebSocket is closed".to_string(),
-            ));
+            return Err(Error::Network("Responses WebSocket is closed".to_string()));
         };
         socket
             .send(Message::Text(text))
             .await
-            .map_err(|error| CoreError::Network(error.to_string()))
+            .map_err(|error| Error::Network(error.to_string()))
     }
 
-    pub async fn recv_text(&self) -> CoreResult<Option<String>> {
+    pub async fn recv_text(&self) -> Result<Option<String>, Error> {
         let mut socket_guard = self.socket.lock().await;
         let Some(socket) = socket_guard.as_mut() else {
             return Ok(None);
@@ -90,27 +88,27 @@ impl ResponsesWebSocketConnection {
             Some(Ok(Message::Text(text))) => Ok(Some(text)),
             Some(Ok(Message::Binary(bytes))) => String::from_utf8(bytes.to_vec())
                 .map(Some)
-                .map_err(|error| CoreError::InvalidResponse(error.to_string())),
+                .map_err(|error| Error::InvalidResponse(error.to_string())),
             Some(Ok(Message::Close(_))) | None => Ok(None),
             Some(Ok(_)) => Ok(None),
-            Some(Err(error)) => Err(CoreError::Network(error.to_string())),
+            Some(Err(error)) => Err(Error::Network(error.to_string())),
         }
     }
 
-    pub async fn close(&self) -> CoreResult<()> {
+    pub async fn close(&self) -> Result<(), Error> {
         let mut socket = self.socket.lock().await;
         if let Some(socket) = socket.as_mut() {
             socket
                 .close(None)
                 .await
-                .map_err(|error| CoreError::Network(error.to_string()))?;
+                .map_err(|error| Error::Network(error.to_string()))?;
         }
         *socket = None;
         Ok(())
     }
 }
 
-pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
+pub(crate) fn resolve_api_key(api_key: Option<&str>) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|value| !value.is_empty())
@@ -120,38 +118,38 @@ pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
                 .ok()
                 .filter(|value| !value.trim().is_empty())
         })
-        .ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
+        .ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
 }
 
 async fn dial_upstream(
     model: &str,
     api_key: &str,
     api_base: Option<&str>,
-) -> CoreResult<ResponsesUpstreamWs> {
+) -> Result<ResponsesUpstreamWs, Error> {
     let url = OPENAI_RESPONSES_WS_CONFIG.complete_websocket_url(api_base, model);
     let mut request = url
         .as_str()
         .into_client_request()
-        .map_err(|error| CoreError::Network(error.to_string()))?;
+        .map_err(|error| Error::Network(error.to_string()))?;
     request.headers_mut().insert(
         AUTHORIZATION,
         HeaderValue::from_str(&format!("Bearer {api_key}"))
-            .map_err(|error| CoreError::Auth(error.to_string()))?,
+            .map_err(|error| Error::Auth(error.to_string()))?,
     );
     let result = tokio::time::timeout(
         Duration::from_secs(DEFAULT_RESPONSES_WS_CONNECT_TIMEOUT_SECS),
         connect_async(request),
     )
     .await
-    .map_err(|_| CoreError::Network("Responses WebSocket connection timed out".to_string()))?;
+    .map_err(|_| Error::Network("Responses WebSocket connection timed out".to_string()))?;
     result
         .map(|(socket, _)| socket)
         .map_err(|error| match error {
-            tokio_tungstenite::tungstenite::Error::Http(response) => CoreError::Http {
+            tokio_tungstenite::tungstenite::Error::Http(response) => Error::Http {
                 status: response.status().as_u16(),
                 body: String::new(),
             },
-            other => CoreError::Network(other.to_string()),
+            other => Error::Network(other.to_string()),
         })
 }
 
@@ -166,7 +164,7 @@ impl ResponsesWebSocketStreaming {
         observe: impl FnMut(&ResponsesWsEvent) + Send,
         client_in: In,
         client_out: Out,
-    ) -> CoreResult<()>
+    ) -> Result<(), Error>
     where
         In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
         Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -193,7 +191,7 @@ pub(crate) async fn splice<In, Out>(
     mut observe: impl FnMut(&ResponsesWsEvent) + Send,
     mut client_in: In,
     mut client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -210,18 +208,18 @@ where
                     .events
                 {
                     let payload = serde_json::to_string(&outbound)
-                        .map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
+                        .map_err(|error| Error::InvalidResponse(error.to_string()))?;
                     upstream_tx.send(Message::Text(payload))
                         .await
-                        .map_err(|error| CoreError::Network(error.to_string()))?;
+                        .map_err(|error| Error::Network(error.to_string()))?;
                 }
             }
             message = upstream_rx.next() => {
                 let Some(message) = message else { break };
-                match message.map_err(|error| CoreError::Network(error.to_string()))? {
+                match message.map_err(|error| Error::Network(error.to_string()))? {
                     Message::Text(text) => {
                         let event = serde_json::from_str::<ResponsesWsEvent>(&text)
-                            .map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
+                            .map_err(|error| Error::InvalidResponse(error.to_string()))?;
                         observe(&event);
                         for outbound in OPENAI_RESPONSES_WS_CONFIG
                             .transform_ws_response(&event, model)?
@@ -229,7 +227,7 @@ where
                         {
                             client_out.send(outbound)
                                 .await
-                                .map_err(|error| CoreError::Network(error.to_string()))?;
+                                .map_err(|error| Error::Network(error.to_string()))?;
                         }
                     }
                     Message::Close(_) => break,
@@ -252,7 +250,7 @@ pub async fn async_responses_websocket<In, Out>(
     mut observe: impl FnMut(&ResponsesWsEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -267,11 +265,11 @@ where
             .events
         {
             let payload = serde_json::to_string(&outbound)
-                .map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
+                .map_err(|error| Error::InvalidResponse(error.to_string()))?;
             upstream_tx
                 .send(Message::Text(payload))
                 .await
-                .map_err(|error| CoreError::Network(error.to_string()))?;
+                .map_err(|error| Error::Network(error.to_string()))?;
         }
     }
     ResponsesWebSocketStreaming::bidirectional_forward(
@@ -296,7 +294,7 @@ pub async fn responses_ws<In, Out>(
     observe: impl FnMut(&ResponsesWsEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -514,7 +512,7 @@ mod tests {
         )
         .await
         .expect_err("status error");
-        assert!(matches!(error, CoreError::Http { status: 401, .. }));
+        assert!(matches!(error, Error::Http { status: 401, .. }));
         server.await.expect("server task");
     }
 
@@ -543,7 +541,7 @@ mod tests {
         )
         .await
         .expect_err("status error");
-        assert!(matches!(error, CoreError::Http { status: 500, .. }));
+        assert!(matches!(error, Error::Http { status: 500, .. }));
         server.await.expect("server task");
     }
 }

litellm-rust/crates/ai-gateway/src/ocr/common_utils.rs

@@ -3,8 +3,7 @@ use std::time::{Duration, Instant};
 
 use base64::Engine;
 use base64::engine::general_purpose::STANDARD as BASE64_STANDARD;
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrProviderConfig;
 use reqwest::Url;
 use serde_json::{Map, Value};
@@ -56,7 +55,7 @@ fn is_azure_document_intelligence_model(model: &str) -> bool {
 
 pub(super) fn string_headers(
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     extra_headers
         .unwrap_or_default()
         .into_iter()
@@ -65,7 +64,7 @@ pub(super) fn string_headers(
                 .as_str()
                 .map(|value| (key.clone(), value.to_string()))
                 .ok_or_else(|| {
-                    CoreError::InvalidRequest(format!(
+                    Error::InvalidRequest(format!(
                         "OCR extra_headers.{key} must be a string, got {}",
                         litellm_core::error::json_type_name(&value)
                     ))
@@ -80,7 +79,7 @@ pub(super) fn has_header(headers: &[(String, String)], name: &str) -> bool {
         .any(|(key, _)| key.eq_ignore_ascii_case(name))
 }
 
-fn document_url_field(document: &Value) -> CoreResult<Option<(&str, &str)>> {
+fn document_url_field(document: &Value) -> Result<Option<(&str, &str)>, Error> {
     let Some(object) = document.as_object() else {
         return Ok(None);
     };
@@ -138,13 +137,13 @@ fn is_blocked_ip(ip: IpAddr) -> bool {
     }
 }
 
-fn blocked_url_error(url: &Url) -> CoreError {
-    CoreError::InvalidRequest(format!(
+fn blocked_url_error(url: &Url) -> Error {
+    Error::InvalidRequest(format!(
         "OCR document URL rejected by SSRF protection: {url}"
     ))
 }
 
-async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
+async fn validate_safe_fetch_url(url: &Url) -> Result<(), Error> {
     if !matches!(url.scheme(), "http" | "https") {
         return Err(blocked_url_error(url));
     }
@@ -162,7 +161,7 @@ async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
         .ok_or_else(|| blocked_url_error(url))?;
     let addresses = tokio::net::lookup_host((host, port))
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     let mut saw_address = false;
     for address in addresses {
         saw_address = true;
@@ -176,25 +175,25 @@ async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
     Ok(())
 }
 
-fn redirect_location(response: &reqwest::Response, url: &Url) -> CoreResult<Url> {
+fn redirect_location(response: &reqwest::Response, url: &Url) -> Result<Url, Error> {
     let location = response
         .headers()
         .get(reqwest::header::LOCATION)
         .and_then(|value| value.to_str().ok())
         .ok_or_else(|| {
-            CoreError::InvalidResponse("OCR document redirect missing Location header".to_string())
+            Error::InvalidResponse("OCR document redirect missing Location header".to_string())
         })?;
     url.join(location)
-        .map_err(|err| CoreError::InvalidResponse(format!("invalid OCR document redirect: {err}")))
+        .map_err(|err| Error::InvalidResponse(format!("invalid OCR document redirect: {err}")))
 }
 
-async fn safe_get_document_url(url: &str) -> CoreResult<(Url, reqwest::Response)> {
+async fn safe_get_document_url(url: &str) -> Result<(Url, reqwest::Response), Error> {
     let client = reqwest::Client::builder()
         .redirect(reqwest::redirect::Policy::none())
         .build()
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     let mut current_url = Url::parse(url)
-        .map_err(|err| CoreError::InvalidRequest(format!("invalid OCR document URL: {err}")))?;
+        .map_err(|err| Error::InvalidRequest(format!("invalid OCR document URL: {err}")))?;
 
     for _ in 0..MAX_SAFE_FETCH_REDIRECTS {
         validate_safe_fetch_url(&current_url).await?;
@@ -202,28 +201,28 @@ async fn safe_get_document_url(url: &str) -> CoreResult<(Url, reqwest::Response)
             .get(current_url.clone())
             .send()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .map_err(|err| Error::Network(err.to_string()))?;
         if !response.status().is_redirection() {
             return Ok((current_url, response));
         }
         current_url = redirect_location(&response, &current_url)?;
     }
 
-    Err(CoreError::InvalidRequest(
+    Err(Error::InvalidRequest(
         "Too many redirects while fetching OCR document URL".to_string(),
     ))
 }
 
-fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> CoreResult<()> {
+fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> Result<(), Error> {
     if max_bytes == 0 {
-        return Err(CoreError::InvalidRequest(format!(
+        return Err(Error::InvalidRequest(format!(
             "OCR document URL download is disabled (MAX_IMAGE_URL_DOWNLOAD_SIZE_MB=0). url={url}"
         )));
     }
     if content_length > max_bytes {
         let size_mb = content_length as f64 / (1024.0 * 1024.0);
         let max_size_mb = max_bytes as f64 / (1024.0 * 1024.0);
-        return Err(CoreError::InvalidRequest(format!(
+        return Err(Error::InvalidRequest(format!(
             "OCR document size ({size_mb:.2}MB) exceeds maximum allowed size ({max_size_mb:.2}MB). url={url}"
         )));
     }
@@ -233,7 +232,7 @@ fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> Core
 async fn read_response_with_limit(
     mut response: reqwest::Response,
     url: &Url,
-) -> CoreResult<Vec<u8>> {
+) -> Result<Vec<u8>, Error> {
     let max_bytes = max_document_download_bytes();
     if let Some(content_length) = response.content_length() {
         enforce_download_size(content_length, max_bytes, url)?;
@@ -246,7 +245,7 @@ async fn read_response_with_limit(
     while let Some(chunk) = response
         .chunk()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?
+        .map_err(|err| Error::Network(err.to_string()))?
     {
         bytes_downloaded += chunk.len() as u64;
         enforce_download_size(bytes_downloaded, max_bytes, url)?;
@@ -255,7 +254,7 @@ async fn read_response_with_limit(
     Ok(bytes)
 }
 
-pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreResult<Value> {
+pub(super) async fn convert_document_url_to_data_uri(document: Value) -> Result<Value, Error> {
     let Some((field, url)) = document_url_field(&document)? else {
         return Ok(document);
     };
@@ -267,7 +266,7 @@ pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreRes
     let status = response.status();
     if !status.is_success() {
         let body = response.text().await.unwrap_or_default();
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&body),
         });
@@ -290,7 +289,7 @@ pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreRes
     let mut transformed = document
         .as_object()
         .cloned()
-        .ok_or_else(|| CoreError::InvalidRequest("OCR document must be an object".to_string()))?;
+        .ok_or_else(|| Error::InvalidRequest("OCR document must be an object".to_string()))?;
     transformed.insert(field.to_string(), Value::String(data_uri));
     Ok(Value::Object(transformed))
 }
@@ -316,11 +315,11 @@ fn retry_after_secs(response: &reqwest::Response) -> u64 {
         .unwrap_or(2)
 }
 
-fn operation_status(response_json: &Value) -> CoreResult<&str> {
+fn operation_status(response_json: &Value) -> Result<&str, Error> {
     let status = response_json
         .get("status")
         .and_then(Value::as_str)
-        .ok_or(CoreError::MissingField("status"))?;
+        .ok_or(Error::MissingField("status"))?;
     match status {
         "succeeded" => Ok("succeeded"),
         "running" | "notStarted" => Ok("running"),
@@ -330,11 +329,11 @@ fn operation_status(response_json: &Value) -> CoreResult<&str> {
                 .and_then(|error| error.get("message"))
                 .and_then(Value::as_str)
                 .unwrap_or("Unknown error");
-            Err(CoreError::InvalidResponse(format!(
+            Err(Error::InvalidResponse(format!(
                 "Azure Document Intelligence analysis failed: {message}"
             )))
         }
-        other => Err(CoreError::InvalidResponse(format!(
+        other => Err(Error::InvalidResponse(format!(
             "Unknown operation status: {other}"
         ))),
     }
@@ -345,9 +344,9 @@ pub(super) async fn poll_document_intelligence(
     original_url: &str,
     headers: &[(String, String)],
     timeout: Option<Duration>,
-) -> CoreResult<Value> {
+) -> Result<Value, Error> {
     if !same_origin(operation_url, original_url) {
-        return Err(CoreError::InvalidResponse(
+        return Err(Error::InvalidResponse(
             "Azure Document Intelligence: rejected cross-origin polling URL".to_string(),
         ));
     }
@@ -358,7 +357,7 @@ pub(super) async fn poll_document_intelligence(
     ));
     loop {
         if start.elapsed() > timeout {
-            return Err(CoreError::Network(format!(
+            return Err(Error::Network(format!(
                 "Azure Document Intelligence operation polling timed out after {} seconds",
                 timeout.as_secs()
             )));
@@ -373,21 +372,21 @@ pub(super) async fn poll_document_intelligence(
         let response = request_builder
             .send()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .map_err(|err| Error::Network(err.to_string()))?;
         let retry_after = retry_after_secs(&response);
         let status = response.status();
         let text = response
             .text()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .map_err(|err| Error::Network(err.to_string()))?;
         if !status.is_success() {
-            return Err(CoreError::Http {
+            return Err(Error::Http {
                 status: status.as_u16(),
                 body: truncate_error_body(&text),
             });
         }
         let response_json: Value = serde_json::from_str(&text).map_err(|err| {
-            CoreError::InvalidResponse(format!("invalid Azure DI poll response JSON: {err}"))
+            Error::InvalidResponse(format!("invalid Azure DI poll response JSON: {err}"))
         })?;
         if operation_status(&response_json)? == "succeeded" {
             return Ok(response_json);
@@ -426,7 +425,7 @@ mod tests {
 
         assert!(matches!(
             error,
-            CoreError::InvalidRequest(message)
+            Error::InvalidRequest(message)
                 if message.contains("SSRF protection")
         ));
     }

litellm-rust/crates/ai-gateway/src/ocr/handler.rs

@@ -1,5 +1,4 @@
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrResponseHandling;
 use serde_json::Value;
 
@@ -7,7 +6,7 @@ use super::common_utils::{poll_document_intelligence, truncate_error_body};
 use super::types::ProviderOcrRequest;
 use crate::client::http_client;
 
-pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> CoreResult<Value> {
+pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Result<Value, Error> {
     let mut request_builder = http_client().post(&request.url).json(&request.body);
     for (key, value) in &request.upstream_headers {
         request_builder = request_builder.header(key, value);
@@ -19,7 +18,7 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
     let response = request_builder
         .send()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     let status = response.status();
     if request.config.response_handling() == OcrResponseHandling::AzureDocumentIntelligencePoll
@@ -31,7 +30,7 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
             .and_then(|value| value.to_str().ok())
             .map(str::to_string)
             .ok_or_else(|| {
-                CoreError::InvalidResponse(
+                Error::InvalidResponse(
                     "Azure Document Intelligence returned 202 but no Operation-Location header found"
                         .to_string(),
                 )
@@ -52,17 +51,17 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
     let text = response
         .text()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
 
     let response_json: Value = serde_json::from_str(&text)
-        .map_err(|err| CoreError::InvalidResponse(format!("invalid OCR response JSON: {err}")))?;
+        .map_err(|err| Error::InvalidResponse(format!("invalid OCR response JSON: {err}")))?;
 
     Ok(request
         .config

litellm-rust/crates/ai-gateway/src/ocr/hooks.rs

@@ -1,11 +1,9 @@
-use std::future::Future;
-use std::pin::Pin;
-
-use litellm_core::CoreResult;
 use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrAuthStrategy;
 use serde_json::{Map, Value, json};
+use std::future::Future;
+use std::pin::Pin;
 
 use super::common_utils::{
     convert_document_url_to_data_uri, has_header, ocr_provider_config, string_headers,
@@ -27,7 +25,7 @@ pub(crate) struct OcrLifecycleHooks {
     request_metadata: RequestMetadata,
 }
 
-type OcrFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
+type OcrFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
 type OcrLogFuture<'a> = Pin<Box<dyn Future<Output = ()> + Send + 'a>>;
 
 impl OcrLifecycleHooks {
@@ -46,7 +44,7 @@ impl OcrLifecycleHooks {
     async fn run_pre_call_guardrails(
         &self,
         request: PreparedOcrRequest,
-    ) -> CoreResult<PreparedOcrRequest> {
+    ) -> Result<PreparedOcrRequest, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(request);
         }
@@ -74,9 +72,9 @@ impl OcrLifecycleHooks {
     async fn prepare_provider_request(
         &self,
         request: PreparedOcrRequest,
-    ) -> CoreResult<ProviderOcrRequest> {
+    ) -> Result<ProviderOcrRequest, Error> {
         let config = ocr_provider_config(&request.custom_llm_provider, &request.model)
-            .ok_or_else(|| CoreError::InvalidProvider(request.custom_llm_provider.clone()))?;
+            .ok_or_else(|| Error::InvalidProvider(request.custom_llm_provider.clone()))?;
         let env_lookup = |key: &str| std::env::var(key).ok();
         let headers = string_headers(request.extra_headers)?;
         let auth_strategy = config.auth_strategy();
@@ -120,7 +118,7 @@ impl OcrLifecycleHooks {
         custom_llm_provider: &str,
         url: &str,
         body: Value,
-    ) -> CoreResult<Value> {
+    ) -> Result<Value, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(body);
         }
@@ -217,7 +215,7 @@ impl CallLifecycleHooks<PreparedOcrRequest, ProviderOcrRequest, Value> for OcrLi
     fn async_log_failure_event<'a>(
         &'a self,
         context: &'a CallLifecycleContext,
-        error: &'a CoreError,
+        error: &'a Error,
         timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a> {
         Box::pin(async move {
@@ -278,19 +276,19 @@ fn guardrail_context(metadata: &RequestMetadata) -> GuardrailContext {
 
 fn parse_ocr_pre_call_guardrail_request(
     request: GuardrailRequest,
-) -> CoreResult<(Value, Map<String, Value>)> {
+) -> Result<(Value, Map<String, Value>), Error> {
     let Value::Object(mut data) = request.data else {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "OCR pre_call guardrail must return an object".to_string(),
         ));
     };
     let document = data.remove("document").ok_or_else(|| {
-        CoreError::InvalidRequest("OCR pre_call guardrail removed document".to_string())
+        Error::InvalidRequest("OCR pre_call guardrail removed document".to_string())
     })?;
     let optional_params = match data.remove("optional_params") {
         Some(Value::Object(params)) => params,
         Some(_) => {
-            return Err(CoreError::InvalidRequest(
+            return Err(Error::InvalidRequest(
                 "OCR pre_call guardrail optional_params must be an object".to_string(),
             ));
         }
@@ -299,33 +297,32 @@ fn parse_ocr_pre_call_guardrail_request(
     Ok((document, optional_params))
 }
 
-fn parse_ocr_during_call_guardrail_request(request: GuardrailRequest) -> CoreResult<Value> {
+fn parse_ocr_during_call_guardrail_request(request: GuardrailRequest) -> Result<Value, Error> {
     let Value::Object(mut data) = request.data else {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "OCR during_call guardrail must return an object".to_string(),
         ));
     };
-    data.remove("body").ok_or_else(|| {
-        CoreError::InvalidRequest("OCR during_call guardrail removed body".to_string())
-    })
+    data.remove("body")
+        .ok_or_else(|| Error::InvalidRequest("OCR during_call guardrail removed body".to_string()))
 }
 
-fn guardrail_error_to_core_error(error: GuardrailError) -> CoreError {
-    CoreError::InvalidRequest(format!("{}: {}", error.kind, error.message))
+fn guardrail_error_to_core_error(error: GuardrailError) -> Error {
+    Error::InvalidRequest(format!("{}: {}", error.kind, error.message))
 }
 
-fn core_error_kind(error: &CoreError) -> &'static str {
+fn core_error_kind(error: &Error) -> &'static str {
     match error {
-        CoreError::Auth(_) => "AuthError",
-        CoreError::InvalidProvider(_) => "InvalidProvider",
-        CoreError::InvalidRequest(_) => "InvalidRequest",
-        CoreError::InvalidType { .. } => "InvalidType",
-        CoreError::MissingField(_) => "MissingField",
-        CoreError::Http { .. } => "HttpError",
-        CoreError::InvalidResponse(_) => "InvalidResponse",
-        CoreError::Network(_) => "NetworkError",
-        CoreError::Connect(_) => "ConnectError",
-        CoreError::Routing(_) => "RoutingError",
-        CoreError::Unsupported(_) => "UnsupportedRequest",
+        Error::Auth(_) => "AuthError",
+        Error::InvalidProvider(_) => "InvalidProvider",
+        Error::InvalidRequest(_) => "InvalidRequest",
+        Error::InvalidType { .. } => "InvalidType",
+        Error::MissingField(_) => "MissingField",
+        Error::Http { .. } => "HttpError",
+        Error::InvalidResponse(_) => "InvalidResponse",
+        Error::Network(_) => "NetworkError",
+        Error::Connect(_) => "ConnectError",
+        Error::Routing(_) => "RoutingError",
+        Error::Unsupported(_) => "UnsupportedRequest",
     }
 }

litellm-rust/crates/ai-gateway/src/ocr/mod.rs

@@ -1,4 +1,4 @@
-use litellm_core::CoreResult;
+use litellm_core::Error;
 use litellm_core::call_lifecycle::CallLifecycle;
 use serde_json::Value;
 
@@ -13,7 +13,7 @@ pub use types::OcrRequest;
 use handler::execute_ocr_provider_call;
 use prepare::{PreparedOcrCall, prepare_ocr_call};
 
-pub async fn ocr(request: OcrRequest<'_>) -> CoreResult<Value> {
+pub async fn ocr(request: OcrRequest<'_>) -> Result<Value, Error> {
     let PreparedOcrCall { request, hooks } = prepare_ocr_call(request);
     CallLifecycle::default()
         .run_request(request, &hooks, execute_ocr_provider_call)

litellm-rust/crates/ai-gateway/src/ocr/tests.rs

@@ -1,7 +1,7 @@
 use std::sync::{Arc, Mutex};
 use std::time::Duration;
 
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrResponseHandling;
 use serde_json::{Map, Value, json};
 use tokio::io::{AsyncReadExt, AsyncWriteExt};
@@ -395,7 +395,7 @@ async fn ocr_lifecycle_runs_failure_hook_on_provider_error() {
     .await
     .expect_err("provider error propagates");
 
-    assert!(matches!(err, CoreError::Http { status: 500, .. }));
+    assert!(matches!(err, Error::Http { status: 500, .. }));
     server.await.expect("server task completes");
     assert_eq!(
         logger.events(),
@@ -439,7 +439,7 @@ async fn ocr_lifecycle_pre_call_block_skips_provider_socket() {
     .await
     .expect_err("guardrail blocks request");
 
-    assert!(matches!(err, CoreError::InvalidRequest(_)));
+    assert!(matches!(err, Error::InvalidRequest(_)));
     assert_eq!(guardrail.events(), vec!["async_pre_call_hook"]);
     assert_eq!(
         logger.events(),
@@ -607,7 +607,7 @@ fn string_headers_rejects_non_string_values() {
     let err = string_headers(Some(headers)).expect_err("non-string header rejected");
     assert_eq!(
         err,
-        CoreError::InvalidRequest(
+        Error::InvalidRequest(
             "OCR extra_headers.x-retry-count must be a string, got number".to_string()
         )
     );

litellm-rust/crates/ai-gateway/src/python/config.rs

@@ -6,33 +6,31 @@
 //! (and recorded in [`crate::gil`]); the realtime hot path never touches Python.
 //!
 //! Compiled only under the `python-config` feature.
-
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::router::{Deployment, Router};
 use pyo3::prelude::*;
 
 use crate::gil;
 
 /// Load the router's `model_list` from `config_path` via the Python reader.
-pub fn load_router_from_config(config_path: &str) -> CoreResult<Router> {
+pub fn load_router_from_config(config_path: &str) -> Result<Router, Error> {
     gil::record_acquisition();
     Python::attach(|py| {
         let model_list = py
             .import("litellm.proxy.read_model_list")
             .and_then(|module| module.getattr("read_model_list"))
             .and_then(|reader| reader.call1((config_path,)))
-            .map_err(|err| CoreError::Routing(format!("read_model_list failed: {err}")))?;
+            .map_err(|err| Error::Routing(format!("read_model_list failed: {err}")))?;
 
         let model_list_json: String = py
             .import("json")
             .and_then(|json| json.getattr("dumps"))
             .and_then(|dumps| dumps.call1((model_list,)))
             .and_then(|encoded| encoded.extract())
-            .map_err(|err| CoreError::Routing(format!("serializing model_list failed: {err}")))?;
+            .map_err(|err| Error::Routing(format!("serializing model_list failed: {err}")))?;
 
         let deployments: Vec<Deployment> = serde_json::from_str(&model_list_json)
-            .map_err(|err| CoreError::Routing(format!("parsing model_list failed: {err}")))?;
+            .map_err(|err| Error::Routing(format!("parsing model_list failed: {err}")))?;
 
         Ok(Router::new(deployments))
     })

litellm-rust/crates/ai-gateway/src/routes/messages/mod.rs

@@ -9,7 +9,7 @@ use axum::http::StatusCode;
 use axum::http::header::{CACHE_CONTROL, CONTENT_TYPE, HeaderMap, HeaderValue};
 use axum::response::{IntoResponse, Response};
 use axum::routing::post;
-use litellm_core::CoreError;
+use litellm_core::Error;
 use serde_json::{Map, Value};
 
 use crate::auth::RequireMasterKey;
@@ -46,7 +46,7 @@ fn stream_response(upstream: reqwest::Response) -> Result<Response, MessagesRout
     let mut response = Response::builder()
         .status(
             StatusCode::from_u16(upstream.status().as_u16()).map_err(|error| {
-                MessagesRouteError(CoreError::InvalidResponse(format!(
+                MessagesRouteError(Error::InvalidResponse(format!(
                     "invalid upstream response status: {error}"
                 )))
             })?,
@@ -58,13 +58,13 @@ fn stream_response(upstream: reqwest::Response) -> Result<Response, MessagesRout
     response
         .body(Body::from_stream(upstream.bytes_stream()))
         .map_err(|error| {
-            MessagesRouteError(CoreError::InvalidResponse(format!(
+            MessagesRouteError(Error::InvalidResponse(format!(
                 "failed to build streaming response: {error}"
             )))
         })
 }
 
-fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>, CoreError> {
+fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>, Error> {
     let forwarded = headers
         .iter()
         .filter(|(name, _)| {
@@ -74,19 +74,19 @@ fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>,
         })
         .map(|(name, value)| {
             let value = value.to_str().map_err(|_| {
-                CoreError::InvalidRequest(format!("invalid value for header {}", name.as_str()))
+                Error::InvalidRequest(format!("invalid value for header {}", name.as_str()))
             })?;
             Ok((name.to_string(), Value::String(value.to_string())))
         })
-        .collect::<Result<Map<_, _>, CoreError>>()?;
+        .collect::<Result<Map<_, _>, Error>>()?;
     Ok((!forwarded.is_empty()).then_some(forwarded))
 }
 
 #[derive(Debug)]
-struct MessagesRouteError(CoreError);
+struct MessagesRouteError(Error);
 
-impl From<CoreError> for MessagesRouteError {
-    fn from(error: CoreError) -> Self {
+impl From<Error> for MessagesRouteError {
+    fn from(error: Error) -> Self {
         Self(error)
     }
 }
@@ -94,28 +94,28 @@ impl From<CoreError> for MessagesRouteError {
 impl IntoResponse for MessagesRouteError {
     fn into_response(self) -> Response {
         let (status, message) = match self.0 {
-            CoreError::InvalidRequest(message) => (StatusCode::BAD_REQUEST, message),
-            CoreError::InvalidProvider(_) | CoreError::Routing(_) => (
+            Error::InvalidRequest(message) => (StatusCode::BAD_REQUEST, message),
+            Error::InvalidProvider(_) | Error::Routing(_) => (
                 StatusCode::NOT_FOUND,
                 "no messages deployment is configured for this model".to_string(),
             ),
-            CoreError::Auth(_) => (
+            Error::Auth(_) => (
                 StatusCode::BAD_GATEWAY,
                 "messages provider authentication failed".to_string(),
             ),
-            CoreError::Http { .. }
-            | CoreError::Network(_)
-            | CoreError::Connect(_)
-            | CoreError::InvalidResponse(_)
-            | CoreError::InvalidType { .. }
-            | CoreError::MissingField(_) => (
+            Error::Http { .. }
+            | Error::Network(_)
+            | Error::Connect(_)
+            | Error::InvalidResponse(_)
+            | Error::InvalidType { .. }
+            | Error::MissingField(_) => (
                 StatusCode::BAD_GATEWAY,
                 "messages provider request failed".to_string(),
             ),
             // The gateway has no Python implementation to decline to, so a
             // request the core cannot serve is reported to the caller. The
             // reason is a fixed internal string, never provider content.
-            CoreError::Unsupported(reason) => (
+            Error::Unsupported(reason) => (
                 StatusCode::BAD_REQUEST,
                 format!("messages request is not supported: {reason}"),
             ),

litellm-rust/crates/ai-gateway/src/routes/messages/service.rs

@@ -1,10 +1,10 @@
 use std::sync::Arc;
 
+use litellm_core::Error;
 use litellm_core::constants::ANTHROPIC_MESSAGES_PROVIDER;
 use litellm_core::messages::types::MessagesRequest;
 use litellm_core::messages::{messages, messages_stream};
 use litellm_core::router::Router;
-use litellm_core::{CoreError, CoreResult};
 use serde_json::{Map, Value};
 
 pub(crate) enum MessagesResponse {
@@ -16,16 +16,16 @@ pub async fn run(
     router: &Arc<Router>,
     body: Value,
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<MessagesResponse> {
+) -> Result<MessagesResponse, Error> {
     let model = body
         .get("model")
         .and_then(Value::as_str)
         .map(str::trim)
         .filter(|model| !model.is_empty())
-        .ok_or_else(|| CoreError::InvalidRequest("messages body requires a model".to_string()))?;
-    let deployment = router.get_available_deployment(model).ok_or_else(|| {
-        CoreError::Routing(format!("no deployment available for model '{model}'"))
-    })?;
+        .ok_or_else(|| Error::InvalidRequest("messages body requires a model".to_string()))?;
+    let deployment = router
+        .get_available_deployment(model)
+        .ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
     let provider_model = deployment.litellm_params.model.as_str();
     let upstream_model = provider_model
         .split_once('/')
@@ -37,7 +37,7 @@ pub async fn run(
     };
     let mut body = body;
     body.as_object_mut()
-        .ok_or_else(|| CoreError::InvalidRequest("messages body must be an object".to_string()))?
+        .ok_or_else(|| Error::InvalidRequest("messages body must be an object".to_string()))?
         .insert(
             "model".to_string(),
             Value::String(upstream_model.to_string()),
@@ -60,6 +60,6 @@ pub async fn run(
     serde_json::to_value(response)
         .map(MessagesResponse::Json)
         .map_err(|err| {
-            CoreError::InvalidResponse(format!("failed to serialize messages response: {err}"))
+            Error::InvalidResponse(format!("failed to serialize messages response: {err}"))
         })
 }

litellm-rust/crates/ai-gateway/src/routes/realtime/service.rs

@@ -11,8 +11,7 @@ use std::time::Duration;
 
 use crate::io::realtime_pool::{RealtimePool, upstream_key};
 use futures_util::{Sink, Stream};
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::realtime::types::RealtimeEvent;
 use litellm_core::router::Router;
 
@@ -29,15 +28,15 @@ pub async fn run<In, Out>(
     observe: impl FnMut(&RealtimeEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,
     <Out as Sink<RealtimeEvent>>::Error: std::fmt::Display,
 {
-    let deployment = router.get_available_deployment(model).ok_or_else(|| {
-        CoreError::Routing(format!("no deployment available for model '{model}'"))
-    })?;
+    let deployment = router
+        .get_available_deployment(model)
+        .ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
     let params = &deployment.litellm_params;
     // Strip a leading `openai/` so the OpenAI-only realtime fn gets the bare model.
     let provider_model = params

litellm-rust/crates/ai-gateway/src/routes/responses/service.rs

@@ -2,13 +2,13 @@ use std::sync::Arc;
 use std::time::Duration;
 
 use futures_util::{Sink, Stream};
+use litellm_core::Error;
 use litellm_core::call_lifecycle::{CallLifecycle, CallLifecycleContext};
 use litellm_core::responses::instrumentation::{
     ResponsesWsCallbackPayload, ResponsesWsInstrumentation, ResponsesWsLogOutcome,
     ResponsesWsMetadata,
 };
 use litellm_core::responses::types::ResponsesWsEvent;
-use litellm_core::{CoreError, CoreResult};
 
 use crate::integrations::custom_logger::{
     CallbackTiming, CallbackValue, CustomLogger, CustomLoggerRunner, LoggingError, ModelCallDetails,
@@ -26,22 +26,22 @@ pub async fn run<In, Out>(
     metadata: RequestMetadata,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
     Out::Error: std::fmt::Display,
 {
-    let deployment = router.get_available_deployment(model).ok_or_else(|| {
-        CoreError::Routing(format!("no deployment available for model '{model}'"))
-    })?;
+    let deployment = router
+        .get_available_deployment(model)
+        .ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
     let params = &deployment.litellm_params;
     let provider_model = params
         .model
         .strip_prefix("openai/")
         .unwrap_or(&params.model);
     if params.model.contains('/') && !params.model.starts_with("openai/") {
-        return Err(CoreError::InvalidProvider(
+        return Err(Error::InvalidProvider(
             "Responses WebSocket route supports OpenAI deployments only".to_string(),
         ));
     }

litellm-rust/crates/core/src/audio_transcription/transformation.rs

@@ -1,7 +1,6 @@
+use crate::Error;
 use serde_json::{Map, Value};
 
-use crate::CoreResult;
-
 use super::types::{AudioTranscriptionRequestData, AudioTranscriptionResponseData};
 
 #[derive(Clone, Debug, PartialEq, Eq)]
@@ -32,13 +31,13 @@ pub trait AudioTranscriptionProviderConfig: Sync {
         model: &str,
         audio: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<AudioTranscriptionRequestData>;
+    ) -> Result<AudioTranscriptionRequestData, Error>;
 
     fn transform_transcription_response(
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<AudioTranscriptionResponseData>;
+    ) -> Result<AudioTranscriptionResponseData, Error>;
 
     fn complete_url(
         &self,
@@ -46,12 +45,12 @@ pub trait AudioTranscriptionProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth_strategy(
         &self,
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<AudioTranscriptionAuth>;
+    ) -> Result<AudioTranscriptionAuth, Error>;
 }

litellm-rust/crates/core/src/call_lifecycle/mod.rs

@@ -1,7 +1,7 @@
 use std::future::Future;
 use std::time::{Instant, SystemTime, UNIX_EPOCH};
 
-use crate::{CoreError, CoreResult};
+use crate::Error;
 
 pub mod types;
 
@@ -11,14 +11,14 @@ pub use types::{
 };
 
 pub trait CallLifecycleHooks<InitialReq, ProviderReq, Resp>: Send + Sync {
-    type PreCallFuture<'a>: Future<Output = CoreResult<InitialReq>> + Send + 'a
+    type PreCallFuture<'a>: Future<Output = Result<InitialReq, Error>> + Send + 'a
     where
         Self: 'a,
         InitialReq: 'a,
         ProviderReq: 'a,
         Resp: 'a;
 
-    type DuringCallFuture<'a>: Future<Output = CoreResult<ProviderReq>> + Send + 'a
+    type DuringCallFuture<'a>: Future<Output = Result<ProviderReq, Error>> + Send + 'a
     where
         Self: 'a,
         InitialReq: 'a,
@@ -56,7 +56,7 @@ pub trait CallLifecycleHooks<InitialReq, ProviderReq, Resp>: Send + Sync {
     fn async_log_failure_event<'a>(
         &'a self,
         context: &'a CallLifecycleContext,
-        error: &'a CoreError,
+        error: &'a Error,
         timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a>;
 }
@@ -86,12 +86,12 @@ impl<'a> CallLifecycle<'a> {
         request: InitialReq,
         hooks: &Hooks,
         provider_call: ProviderCall,
-    ) -> CoreResult<Resp>
+    ) -> Result<Resp, Error>
     where
         InitialReq: CallLifecycleRequest,
         Hooks: CallLifecycleHooks<InitialReq, ProviderReq, Resp>,
         ProviderCall: FnOnce(ProviderReq) -> ProviderFuture,
-        ProviderFuture: Future<Output = CoreResult<Resp>>,
+        ProviderFuture: Future<Output = Result<Resp, Error>>,
     {
         let context = request.lifecycle_context();
         self.run(context, request, hooks, provider_call).await
@@ -103,11 +103,11 @@ impl<'a> CallLifecycle<'a> {
         request: InitialReq,
         hooks: &Hooks,
         provider_call: ProviderCall,
-    ) -> CoreResult<Resp>
+    ) -> Result<Resp, Error>
     where
         Hooks: CallLifecycleHooks<InitialReq, ProviderReq, Resp>,
         ProviderCall: FnOnce(ProviderReq) -> ProviderFuture,
-        ProviderFuture: Future<Output = CoreResult<Resp>>,
+        ProviderFuture: Future<Output = Result<Resp, Error>>,
     {
         let call_start = epoch_seconds();
         let mut phases = Vec::new();
@@ -166,7 +166,7 @@ impl<'a> CallLifecycle<'a> {
         &self,
         context: &CallLifecycleContext,
         hooks: &Hooks,
-        error: &CoreError,
+        error: &Error,
         call_start: f64,
         phases: &mut Vec<CallLifecyclePhaseTiming>,
     ) where
@@ -251,8 +251,8 @@ mod tests {
     }
 
     impl CallLifecycleHooks<String, String, String> for RecordingHooks {
-        type PreCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
-        type DuringCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
+        type PreCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
+        type DuringCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
         type SuccessFuture<'a> = BoxFuture<'a, ()>;
         type FailureFuture<'a> = BoxFuture<'a, ()>;
 
@@ -294,7 +294,7 @@ mod tests {
         fn async_log_failure_event<'a>(
             &'a self,
             _context: &'a CallLifecycleContext,
-            _error: &'a CoreError,
+            _error: &'a Error,
             _timing: &'a CallLifecycleTiming,
         ) -> Self::FailureFuture<'a> {
             Box::pin(async move {
@@ -304,8 +304,8 @@ mod tests {
     }
 
     impl CallLifecycleHooks<RecordingRequest, String, String> for RecordingHooks {
-        type PreCallFuture<'a> = BoxFuture<'a, CoreResult<RecordingRequest>>;
-        type DuringCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
+        type PreCallFuture<'a> = BoxFuture<'a, Result<RecordingRequest, Error>>;
+        type DuringCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
         type SuccessFuture<'a> = BoxFuture<'a, ()>;
         type FailureFuture<'a> = BoxFuture<'a, ()>;
 
@@ -345,7 +345,7 @@ mod tests {
         fn async_log_failure_event<'a>(
             &'a self,
             _context: &'a CallLifecycleContext,
-            _error: &'a CoreError,
+            _error: &'a Error,
             _timing: &'a CallLifecycleTiming,
         ) -> Self::FailureFuture<'a> {
             Box::pin(async move {
@@ -383,13 +383,13 @@ mod tests {
                 "request".to_string(),
                 &hooks,
                 |_request| async move {
-                    Err::<String, CoreError>(CoreError::Network("provider down".to_string()))
+                    Err::<String, Error>(Error::Network("provider down".to_string()))
                 },
             )
             .await
             .expect_err("call fails");
 
-        assert_eq!(error, CoreError::Network("provider down".to_string()));
+        assert_eq!(error, Error::Network("provider down".to_string()));
         assert_eq!(hooks.events(), vec!["pre_call", "during_call", "failure"]);
     }
 

litellm-rust/crates/core/src/chat_completions/common_utils.rs

@@ -1,8 +1,7 @@
-use serde_json::{Map, Value};
-
-use crate::error::CoreResult;
+use crate::Error;
 use crate::http_utils::string_headers as shared_string_headers;
 use crate::providers::anthropic::chat_completions::transformation::ANTHROPIC_CHAT_COMPLETIONS_CONFIG;
+use serde_json::{Map, Value};
 
 use super::transformation::ChatCompletionsProviderConfig;
 
@@ -23,6 +22,6 @@ pub(super) fn chat_completions_provider_config(
 
 pub(super) fn string_headers(
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     shared_string_headers(HEADER_CONTEXT, extra_headers)
 }

litellm-rust/crates/core/src/chat_completions/handler.rs

@@ -1,6 +1,6 @@
 use serde_json::Value;
 
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::http_utils::truncate_error_body;
 
 use super::client::http_client;
@@ -11,9 +11,9 @@ use super::types::{
 
 pub(super) async fn execute_chat_completions_provider_call(
     request: ProviderChatCompletionsRequest,
-) -> CoreResult<ChatCompletionsResponse> {
+) -> Result<ChatCompletionsResponse, Error> {
     let body = serde_json::to_vec(&request.body).map_err(|err| {
-        CoreError::InvalidRequest(format!(
+        Error::InvalidRequest(format!(
             "failed to serialize chat completions request: {err}"
         ))
     })?;
@@ -32,9 +32,9 @@ pub(super) async fn execute_chat_completions_provider_call(
         // so the host can still serve it. Everything else here, a timeout
         // above all, may have reached the provider and been answered.
         if err.is_connect() || err.is_builder() {
-            CoreError::Connect(err.to_string())
+            Error::Connect(err.to_string())
         } else {
-            CoreError::Network(err.to_string())
+            Error::Network(err.to_string())
         }
     })?;
 
@@ -42,17 +42,17 @@ pub(super) async fn execute_chat_completions_provider_call(
     let text = response
         .text()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
 
     let body: Value = serde_json::from_str(&text).map_err(|err| {
-        CoreError::InvalidResponse(format!("invalid chat completions response JSON: {err}"))
+        Error::InvalidResponse(format!("invalid chat completions response JSON: {err}"))
     })?;
     request
         .config
@@ -69,10 +69,10 @@ pub(super) async fn execute_chat_completions_provider_call(
 /// second kind has already been billed, and a host that keeps a reference
 /// implementation must not retry those, so collapse them to one variant that
 /// can only mean the provider was already called.
-pub(super) fn as_response_error(err: CoreError) -> CoreError {
+pub(super) fn as_response_error(err: Error) -> Error {
     match err {
-        already @ (CoreError::InvalidResponse(_) | CoreError::Http { .. }) => already,
-        other => CoreError::InvalidResponse(other.to_string()),
+        already @ (Error::InvalidResponse(_) | Error::Http { .. }) => already,
+        other => Error::InvalidResponse(other.to_string()),
     }
 }
 
@@ -80,7 +80,7 @@ pub(super) fn as_response_error(err: CoreError) -> CoreError {
 pub(super) async fn signed_headers(
     request: &ProviderChatCompletionsRequest,
     body: &[u8],
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     use std::collections::BTreeMap;
     use std::time::SystemTime;
 
@@ -101,7 +101,7 @@ pub(super) async fn signed_headers(
         .iter()
         .any(|(name, _)| is_sigv4_computed_header(name))
     {
-        return Err(CoreError::Unsupported(
+        return Err(Error::Unsupported(
             "request forwards a header AWS SigV4 computes",
         ));
     }
@@ -137,9 +137,9 @@ pub(super) async fn signed_headers(
 pub(super) async fn signed_headers(
     request: &ProviderChatCompletionsRequest,
     _body: &[u8],
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     match &request.auth {
-        ChatCompletionsAuth::AwsSigV4 { .. } => Err(CoreError::Unsupported(
+        ChatCompletionsAuth::AwsSigV4 { .. } => Err(Error::Unsupported(
             "AWS SigV4 requires the bedrock-auth feature",
         )),
         _ => Ok(request.upstream_headers.clone()),

litellm-rust/crates/core/src/chat_completions/mod.rs

@@ -6,6 +6,7 @@
 //! credentials, and it resolves the provider, translates the conversation,
 //! calls the provider, and returns a typed OpenAI-shaped response.
 
+use crate::Error;
 mod client;
 mod common_utils;
 pub mod conversation;
@@ -17,15 +18,13 @@ pub mod types;
 
 use serde_json::{Map, Value};
 
-use crate::error::CoreResult;
-
 use handler::execute_chat_completions_provider_call;
 use prepare::{parse_messages, prepare_chat_completions_call, resolve_provider_config};
 use types::{ChatCompletionsRequest, ChatCompletionsResponse};
 
 pub async fn chat_completions(
     request: ChatCompletionsRequest<'_>,
-) -> CoreResult<ChatCompletionsResponse> {
+) -> Result<ChatCompletionsResponse, Error> {
     execute_chat_completions_provider_call(prepare_chat_completions_call(request)?).await
 }
 

litellm-rust/crates/core/src/chat_completions/prepare.rs

@@ -1,6 +1,6 @@
 use serde_json::Value;
 
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::http_utils::has_header;
 use crate::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
 
@@ -11,7 +11,7 @@ use super::types::{ChatCompletionsRequest, ChatMessage, ProviderChatCompletionsR
 pub(super) fn resolve_provider_config<'a>(
     model: &'a str,
     custom_llm_provider: Option<&'a str>,
-) -> CoreResult<(String, &'static dyn ChatCompletionsProviderConfig)> {
+) -> Result<(String, &'static dyn ChatCompletionsProviderConfig), Error> {
     let provider_info = get_custom_llm_provider(model, custom_llm_provider)
         .or_else(|| {
             custom_llm_provider.map(|provider| CustomLlmProvider {
@@ -20,35 +20,34 @@ pub(super) fn resolve_provider_config<'a>(
             })
         })
         .ok_or_else(|| {
-            CoreError::InvalidProvider(
+            Error::InvalidProvider(
                 "unable to resolve custom_llm_provider for chat completions request".to_string(),
             )
         })?;
     let config = chat_completions_provider_config(provider_info.custom_llm_provider)
-        .ok_or_else(|| CoreError::InvalidProvider(provider_info.custom_llm_provider.to_string()))?;
+        .ok_or_else(|| Error::InvalidProvider(provider_info.custom_llm_provider.to_string()))?;
     Ok((provider_info.model.to_string(), config))
 }
 
-pub(super) fn parse_messages(messages: Value) -> CoreResult<Vec<ChatMessage>> {
-    serde_json::from_value(messages).map_err(|err| {
-        CoreError::InvalidRequest(format!("invalid chat completions messages: {err}"))
-    })
+pub(super) fn parse_messages(messages: Value) -> Result<Vec<ChatMessage>, Error> {
+    serde_json::from_value(messages)
+        .map_err(|err| Error::InvalidRequest(format!("invalid chat completions messages: {err}")))
 }
 
 pub(super) fn prepare_chat_completions_call(
     request: ChatCompletionsRequest<'_>,
-) -> CoreResult<ProviderChatCompletionsRequest> {
+) -> Result<ProviderChatCompletionsRequest, Error> {
     let (model, config) = resolve_provider_config(request.model, request.custom_llm_provider)?;
     let env_lookup = |key: &str| std::env::var(key).ok();
 
     let messages = parse_messages(request.messages)?;
     if messages.is_empty() {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "chat completions requires at least one message".to_string(),
         ));
     }
     if let Some(reason) = config.unsupported_reason(&messages, &request.optional_params) {
-        return Err(CoreError::Unsupported(reason.0));
+        return Err(Error::Unsupported(reason.0));
     }
 
     let mut headers = string_headers(request.extra_headers)?;

litellm-rust/crates/core/src/chat_completions/tests.rs

@@ -1,6 +1,6 @@
 use serde_json::{Map, Value, json};
 
-use crate::error::CoreError;
+use crate::error::Error;
 
 use super::prepare::prepare_chat_completions_call;
 use super::transformation::ChatCompletionsAuth;
@@ -29,7 +29,7 @@ fn request<'a>(
 
 /// `ProviderChatCompletionsRequest` deliberately has no `Debug` (its headers
 /// carry resolved credentials), so unwrap the failure case by hand.
-fn decline(request: ChatCompletionsRequest<'_>) -> CoreError {
+fn decline(request: ChatCompletionsRequest<'_>) -> Error {
     match prepare_chat_completions_call(request) {
         Err(error) => error,
         Ok(prepared) => panic!("expected a decline, prepared a call to {}", prepared.url),
@@ -196,7 +196,7 @@ fn declines_an_unsupported_request_before_resolving_credentials() {
     call.api_key = None;
     // No api_key is set and no env is consulted: the gate must run first, so the
     // error is the decline rather than a missing-credential error.
-    assert_eq!(decline(call), CoreError::Unsupported("streaming"));
+    assert_eq!(decline(call), Error::Unsupported("streaming"));
 }
 
 #[test]
@@ -208,7 +208,7 @@ fn rejects_an_unknown_provider() {
             json!([{"role": "user", "content": "hi"}]),
             json!({}),
         )),
-        CoreError::InvalidProvider("openai".to_string())
+        Error::InvalidProvider("openai".to_string())
     );
 }
 
@@ -221,7 +221,7 @@ fn rejects_a_model_with_no_resolvable_provider() {
             json!([{"role": "user", "content": "hi"}]),
             json!({}),
         )),
-        CoreError::InvalidProvider(_)
+        Error::InvalidProvider(_)
     ));
 }
 
@@ -234,7 +234,7 @@ fn rejects_an_empty_or_malformed_message_list() {
             json!([]),
             json!({}),
         )),
-        CoreError::InvalidRequest("chat completions requires at least one message".to_string())
+        Error::InvalidRequest("chat completions requires at least one message".to_string())
     );
     assert!(matches!(
         decline(request(
@@ -243,7 +243,7 @@ fn rejects_an_empty_or_malformed_message_list() {
             json!("not a list"),
             json!({}),
         )),
-        CoreError::InvalidRequest(_)
+        Error::InvalidRequest(_)
     ));
 }
 
@@ -258,7 +258,7 @@ fn rejects_non_string_extra_headers() {
     call.extra_headers = Some(Map::from_iter([("x-trace".to_string(), json!(7))]));
     assert_eq!(
         decline(call),
-        CoreError::InvalidRequest(
+        Error::InvalidRequest(
             "chat completions extra_headers.x-trace must be a string, got number".to_string()
         )
     );
@@ -374,7 +374,7 @@ async fn a_forwarded_header_the_signer_computes_declines_to_python() {
             .await
             .expect_err("{forwarded} should decline instead of being signed");
         assert!(
-            matches!(error, CoreError::Unsupported(_)),
+            matches!(error, Error::Unsupported(_)),
             "{forwarded} declined as {error:?}, which the host would not fall back on"
         );
     }
@@ -727,7 +727,7 @@ mod round_trip {
         .expect_err("response cannot be normalized");
         handle.await.expect("server task");
         assert!(
-            matches!(err, CoreError::InvalidResponse(_)),
+            matches!(err, Error::InvalidResponse(_)),
             "expected a post-send error, got {err:?}"
         );
     }
@@ -745,7 +745,7 @@ mod round_trip {
         .expect_err("response cannot be normalized");
         handle.await.expect("server task");
         assert!(
-            matches!(err, CoreError::InvalidResponse(_)),
+            matches!(err, Error::InvalidResponse(_)),
             "expected a post-send error, got {err:?}"
         );
     }
@@ -763,7 +763,7 @@ mod round_trip {
         .expect_err("upstream rejects");
         handle.await.expect("server task");
         assert!(
-            matches!(err, CoreError::Http { status: 429, .. }),
+            matches!(err, Error::Http { status: 429, .. }),
             "expected a 429, got {err:?}"
         );
     }
@@ -787,7 +787,7 @@ mod round_trip {
         .await
         .expect_err("nothing is listening");
         assert!(
-            matches!(err, CoreError::Connect(_)),
+            matches!(err, Error::Connect(_)),
             "expected a pre-send connect failure, got {err:?}"
         );
     }
@@ -797,24 +797,24 @@ mod round_trip {
         use crate::chat_completions::handler::as_response_error;
 
         for original in [
-            CoreError::MissingField("usage"),
-            CoreError::Unsupported("non-text response content block"),
-            CoreError::InvalidRequest("whatever".to_string()),
-            CoreError::Auth("whatever".to_string()),
+            Error::MissingField("usage"),
+            Error::Unsupported("non-text response content block"),
+            Error::InvalidRequest("whatever".to_string()),
+            Error::Auth("whatever".to_string()),
         ] {
             let label = format!("{original:?}");
             assert!(
-                matches!(as_response_error(original), CoreError::InvalidResponse(_)),
+                matches!(as_response_error(original), Error::InvalidResponse(_)),
                 "{label} must not stay retryable once the provider has answered"
             );
         }
         // An upstream status is already unambiguous, so it survives intact.
         assert!(matches!(
-            as_response_error(CoreError::Http {
+            as_response_error(Error::Http {
                 status: 500,
                 body: "boom".to_string()
             }),
-            CoreError::Http { status: 500, .. }
+            Error::Http { status: 500, .. }
         ));
     }
 }

litellm-rust/crates/core/src/chat_completions/transformation.rs

@@ -1,7 +1,6 @@
+use crate::Error;
 use serde_json::{Map, Value};
 
-use crate::error::CoreResult;
-
 use super::types::{
     ChatCompletionsResponse, ChatMessage, ChatMessageContent, ProviderChatRequestData,
     ProviderChatResponseData,
@@ -39,7 +38,7 @@ pub trait ChatCompletionsProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth(
         &self,
@@ -47,7 +46,7 @@ pub trait ChatCompletionsProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<ChatCompletionsAuth>;
+    ) -> Result<ChatCompletionsAuth, Error>;
 
     fn default_headers(&self) -> &'static [(&'static str, &'static str)] {
         &[("content-type", "application/json")]
@@ -91,13 +90,13 @@ pub trait ChatCompletionsProviderConfig: Sync {
         model: &str,
         messages: Vec<ChatMessage>,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<ProviderChatRequestData>;
+    ) -> Result<ProviderChatRequestData, Error>;
 
     fn transform_response(
         &self,
         model: &str,
         response: ProviderChatResponseData,
-    ) -> CoreResult<ChatCompletionsResponse>;
+    ) -> Result<ChatCompletionsResponse, Error>;
 }
 
 pub fn unsupported_param(

litellm-rust/crates/core/src/error.rs

@@ -1,9 +1,7 @@
-use thiserror::Error;
+use thiserror::Error as ThisError;
 
-pub type CoreResult<T> = Result<T, CoreError>;
-
-#[derive(Debug, Error, PartialEq, Eq)]
-pub enum CoreError {
+#[derive(Debug, ThisError, PartialEq, Eq)]
+pub enum Error {
     #[error("expected {expected}, got {actual}")]
     InvalidType {
         expected: &'static str,

litellm-rust/crates/core/src/http_utils.rs

@@ -3,7 +3,7 @@
 use serde_json::{Map, Value};
 
 use crate::constants::UPSTREAM_ERROR_BODY_MAX_CHARS;
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 
 /// Bound an upstream error body before it crosses a host boundary, so provider
 /// bodies stay data-minimized.
@@ -18,7 +18,7 @@ pub fn truncate_error_body(body: &str) -> String {
 pub fn string_headers(
     context: &'static str,
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     extra_headers
         .unwrap_or_default()
         .into_iter()
@@ -27,7 +27,7 @@ pub fn string_headers(
                 .as_str()
                 .map(|value| (key.clone(), value.to_string()))
                 .ok_or_else(|| {
-                    CoreError::InvalidRequest(format!(
+                    Error::InvalidRequest(format!(
                         "{context} extra_headers.{key} must be a string, got {}",
                         json_type_name(&value)
                     ))
@@ -81,7 +81,7 @@ mod tests {
         let err = string_headers("chat completions", Some(headers)).expect_err("non-string value");
         assert_eq!(
             err,
-            CoreError::InvalidRequest(
+            Error::InvalidRequest(
                 "chat completions extra_headers.x-trace must be a string, got number".to_string()
             )
         );

litellm-rust/crates/core/src/lib.rs

@@ -13,4 +13,4 @@ pub mod responses;
 pub mod router;
 pub mod routing_utils;
 
-pub use error::{CoreError, CoreResult};
+pub use error::Error;

litellm-rust/crates/core/src/messages/common_utils.rs

@@ -1,9 +1,8 @@
-use serde_json::{Map, Value};
-
-use crate::error::CoreResult;
+use crate::Error;
 use crate::http_utils::string_headers as shared_string_headers;
 use crate::providers::anthropic::messages::transformation::ANTHROPIC_MESSAGES_CONFIG;
 use crate::providers::azure_ai::messages::transformation::AZURE_ANTHROPIC_MESSAGES_CONFIG;
+use serde_json::{Map, Value};
 
 use super::transformation::AnthropicMessagesProviderConfig;
 
@@ -23,6 +22,6 @@ pub(super) fn messages_provider_config(
 
 pub(super) fn string_headers(
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     shared_string_headers(HEADER_CONTEXT, extra_headers)
 }

litellm-rust/crates/core/src/messages/handler.rs

@@ -1,5 +1,5 @@
 use crate::constants::ANTHROPIC_MESSAGES_PROVIDER;
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 
 use super::client::http_client;
 use super::common_utils::truncate_error_body;
@@ -7,7 +7,7 @@ use super::types::{AnthropicMessagesResponse, ProviderMessagesRequest};
 
 pub(super) async fn execute_messages_provider_call(
     request: ProviderMessagesRequest,
-) -> CoreResult<AnthropicMessagesResponse> {
+) -> Result<AnthropicMessagesResponse, Error> {
     let mut request_builder = http_client().post(&request.url).json(&request.body);
     for (key, value) in &request.upstream_headers {
         request_builder = request_builder.header(key, value);
@@ -19,32 +19,31 @@ pub(super) async fn execute_messages_provider_call(
     let response = request_builder
         .send()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     let status = response.status();
     let text = response
         .text()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
 
-    let response = serde_json::from_str(&text).map_err(|err| {
-        CoreError::InvalidResponse(format!("invalid messages response JSON: {err}"))
-    })?;
+    let response = serde_json::from_str(&text)
+        .map_err(|err| Error::InvalidResponse(format!("invalid messages response JSON: {err}")))?;
     request.config.transform_response(&request.model, response)
 }
 
 pub(super) async fn execute_messages_provider_stream(
     request: ProviderMessagesRequest,
-) -> CoreResult<reqwest::Response> {
+) -> Result<reqwest::Response, Error> {
     if request.provider != ANTHROPIC_MESSAGES_PROVIDER {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "streaming messages is not supported for this provider".to_string(),
         ));
     }
@@ -60,14 +59,14 @@ pub(super) async fn execute_messages_provider_stream(
     let response = request_builder
         .send()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     let status = response.status();
     if !status.is_success() {
         let text = response
             .text()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
-        return Err(CoreError::Http {
+            .map_err(|err| Error::Network(err.to_string()))?;
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });

litellm-rust/crates/core/src/messages/mod.rs

@@ -7,6 +7,7 @@
 //! is the streaming variant; it hands the raw upstream response back so a host
 //! can splice the event stream to its own caller.
 
+use crate::Error;
 mod client;
 mod common_utils;
 mod handler;
@@ -14,17 +15,15 @@ mod prepare;
 pub mod transformation;
 pub mod types;
 
-use crate::error::CoreResult;
-
 use handler::{execute_messages_provider_call, execute_messages_provider_stream};
 use prepare::prepare_messages_call;
 use types::{AnthropicMessagesResponse, MessagesRequest};
 
-pub async fn messages(request: MessagesRequest<'_>) -> CoreResult<AnthropicMessagesResponse> {
+pub async fn messages(request: MessagesRequest<'_>) -> Result<AnthropicMessagesResponse, Error> {
     execute_messages_provider_call(prepare_messages_call(request)?).await
 }
 
-pub async fn messages_stream(request: MessagesRequest<'_>) -> CoreResult<reqwest::Response> {
+pub async fn messages_stream(request: MessagesRequest<'_>) -> Result<reqwest::Response, Error> {
     execute_messages_provider_stream(prepare_messages_call(request)?).await
 }
 

litellm-rust/crates/core/src/messages/prepare.rs

@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
 
 use super::common_utils::{has_bearer_auth, has_header, messages_provider_config, string_headers};
@@ -7,7 +7,7 @@ use super::types::{MessagesRequest, ProviderMessagesRequest};
 
 pub(super) fn prepare_messages_call(
     request: MessagesRequest<'_>,
-) -> CoreResult<ProviderMessagesRequest> {
+) -> Result<ProviderMessagesRequest, Error> {
     let provider_info = get_custom_llm_provider(request.model, request.custom_llm_provider)
         .or_else(|| {
             request
@@ -18,7 +18,7 @@ pub(super) fn prepare_messages_call(
                 })
         })
         .ok_or_else(|| {
-            CoreError::InvalidProvider(
+            Error::InvalidProvider(
                 "unable to resolve custom_llm_provider for messages request".to_string(),
             )
         })?;
@@ -26,7 +26,7 @@ pub(super) fn prepare_messages_call(
     let provider = provider_info.custom_llm_provider;
 
     let config = messages_provider_config(provider)
-        .ok_or_else(|| CoreError::InvalidProvider(provider.to_string()))?;
+        .ok_or_else(|| Error::InvalidProvider(provider.to_string()))?;
     let env_lookup = |key: &str| std::env::var(key).ok();
 
     let mut headers = string_headers(request.extra_headers)?;
@@ -53,11 +53,11 @@ pub(super) fn prepare_messages_call(
 
     let url = config.complete_url(request.api_base, &model, &env_lookup)?;
     let typed_request = serde_json::from_value(request.body).map_err(|err| {
-        CoreError::InvalidRequest(format!("invalid Anthropic messages request: {err}"))
+        Error::InvalidRequest(format!("invalid Anthropic messages request: {err}"))
     })?;
     let transformed = config.transform_request(typed_request)?;
     let body = serde_json::to_value(transformed).map_err(|err| {
-        CoreError::InvalidRequest(format!(
+        Error::InvalidRequest(format!(
             "failed to serialize Anthropic messages request: {err}"
         ))
     })?;

litellm-rust/crates/core/src/messages/tests.rs

@@ -4,7 +4,7 @@ use serde_json::{Map, Value, json};
 use tokio::io::{AsyncReadExt, AsyncWriteExt};
 use tokio::net::{TcpListener, TcpStream};
 
-use crate::error::CoreError;
+use crate::error::Error;
 
 use super::common_utils::{
     has_bearer_auth, has_header, messages_provider_config, string_headers, truncate_error_body,
@@ -77,7 +77,7 @@ fn truncate_error_body_caps_long_payloads() {
 fn string_headers_rejects_non_string_values() {
     let headers = json!({"x-count": 3}).as_object().unwrap().clone();
     let err = string_headers(Some(headers)).expect_err("non-string header rejected");
-    assert!(matches!(err, CoreError::InvalidRequest(_)));
+    assert!(matches!(err, Error::InvalidRequest(_)));
 }
 
 #[test]
@@ -341,7 +341,7 @@ async fn messages_requires_auth_when_no_key_and_no_header() {
     .await
     .expect_err("missing auth errors");
 
-    assert!(matches!(err, CoreError::Auth(_)));
+    assert!(matches!(err, Error::Auth(_)));
 }
 
 #[tokio::test]
@@ -420,7 +420,7 @@ async fn messages_maps_provider_error_status_to_http_error() {
     .await
     .expect_err("provider error propagates");
 
-    assert!(matches!(err, CoreError::Http { status: 401, .. }));
+    assert!(matches!(err, Error::Http { status: 401, .. }));
 }
 
 #[tokio::test]
@@ -437,5 +437,5 @@ async fn messages_rejects_unsupported_provider() {
     .await
     .expect_err("unsupported provider errors");
 
-    assert!(matches!(err, CoreError::InvalidProvider(provider) if provider == "openai"));
+    assert!(matches!(err, Error::InvalidProvider(provider) if provider == "openai"));
 }

litellm-rust/crates/core/src/messages/transformation.rs

@@ -1,6 +1,5 @@
-use crate::error::CoreResult;
-
 use super::types::{AnthropicMessagesRequest, AnthropicMessagesResponse};
+use crate::Error;
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
 pub enum MessagesAuthStrategy {
@@ -23,13 +22,13 @@ pub trait AnthropicMessagesProviderConfig: Sync {
         api_base: Option<&str>,
         model: &str,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn resolve_api_key(
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth_strategy(&self) -> MessagesAuthStrategy {
         MessagesAuthStrategy::Header("x-api-key")
@@ -49,7 +48,7 @@ pub trait AnthropicMessagesProviderConfig: Sync {
     fn transform_request(
         &self,
         request: AnthropicMessagesRequest,
-    ) -> CoreResult<AnthropicMessagesRequest> {
+    ) -> Result<AnthropicMessagesRequest, Error> {
         Ok(request)
     }
 
@@ -57,7 +56,7 @@ pub trait AnthropicMessagesProviderConfig: Sync {
         &self,
         _model: &str,
         response: AnthropicMessagesResponse,
-    ) -> CoreResult<AnthropicMessagesResponse> {
+    ) -> Result<AnthropicMessagesResponse, Error> {
         Ok(response)
     }
 }

litellm-rust/crates/core/src/ocr/transformation.rs

@@ -1,7 +1,6 @@
+use crate::Error;
 use serde_json::{Map, Value};
 
-use crate::CoreResult;
-
 use super::types::{OcrRequestData, OcrResponseData};
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
@@ -43,13 +42,13 @@ pub trait OcrProviderConfig: Sync {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData>;
+    ) -> Result<OcrRequestData, Error>;
 
     fn transform_ocr_response(
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData>;
+    ) -> Result<OcrResponseData, Error>;
 
     fn complete_url(
         &self,
@@ -57,13 +56,13 @@ pub trait OcrProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn resolve_api_key(
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth_strategy(&self) -> OcrAuthStrategy {
         OcrAuthStrategy::Bearer

litellm-rust/crates/core/src/providers/anthropic/chat_completions/tests.rs

@@ -1,4 +1,5 @@
 use super::*;
+use crate::Error;
 use serde_json::json;
 
 fn messages(value: Value) -> Vec<ChatMessage> {
@@ -19,7 +20,7 @@ fn transform(model: &str, msgs: Value, opts: Value) -> Value {
         .body
 }
 
-fn transform_response(body: Value) -> CoreResult<ChatCompletionsResponse> {
+fn transform_response(body: Value) -> Result<ChatCompletionsResponse, Error> {
     ANTHROPIC_CHAT_COMPLETIONS_CONFIG
         .transform_response("claude-sonnet-4-5", ProviderChatResponseData { body })
 }
@@ -390,29 +391,26 @@ fn declines_a_response_carrying_a_non_text_block() {
         "usage": {"input_tokens": 1, "output_tokens": 1}
     }))
     .expect_err("non-text block");
-    assert_eq!(
-        err,
-        CoreError::Unsupported("non-text response content block")
-    );
+    assert_eq!(err, Error::Unsupported("non-text response content block"));
 }
 
 #[test]
 fn errors_on_a_response_missing_required_fields() {
     assert_eq!(
         transform_response(json!("nope")).expect_err("not an object"),
-        CoreError::InvalidResponse("messages response is not an object".to_string())
+        Error::InvalidResponse("messages response is not an object".to_string())
     );
     assert_eq!(
         transform_response(json!({"model": "m", "usage": {}})).expect_err("no content"),
-        CoreError::MissingField("content")
+        Error::MissingField("content")
     );
     assert_eq!(
         transform_response(json!({"model": "m", "content": []})).expect_err("no usage"),
-        CoreError::MissingField("usage")
+        Error::MissingField("usage")
     );
     assert_eq!(
         transform_response(json!({"content": [], "usage": {}})).expect_err("no model"),
-        CoreError::MissingField("model")
+        Error::MissingField("model")
     );
 }
 

litellm-rust/crates/core/src/providers/anthropic/chat_completions/transformation.rs

@@ -10,7 +10,7 @@ use crate::chat_completions::types::{
     ProviderChatRequestData, ProviderChatResponseData,
 };
 use crate::constants::ANTHROPIC_OAUTH_TOKEN_PREFIX;
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::providers::anthropic::messages::transformation::{
     complete_anthropic_url, resolve_anthropic_api_key,
 };
@@ -74,7 +74,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         Ok(complete_anthropic_url(api_base, env_lookup))
     }
 
@@ -84,7 +84,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<ChatCompletionsAuth> {
+    ) -> Result<ChatCompletionsAuth, Error> {
         Ok(ChatCompletionsAuth::Header {
             name: "x-api-key",
             value: resolve_anthropic_api_key(api_key, env_lookup)?,
@@ -137,7 +137,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         model: &str,
         messages: Vec<ChatMessage>,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<ProviderChatRequestData> {
+    ) -> Result<ProviderChatRequestData, Error> {
         Ok(ProviderChatRequestData {
             body: anthropic_body(model, &build_conversation(&messages), optional_params),
         })
@@ -147,15 +147,16 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         &self,
         _model: &str,
         response: ProviderChatResponseData,
-    ) -> CoreResult<ChatCompletionsResponse> {
-        let body = response.body.as_object().ok_or_else(|| {
-            CoreError::InvalidResponse("messages response is not an object".into())
-        })?;
+    ) -> Result<ChatCompletionsResponse, Error> {
+        let body = response
+            .body
+            .as_object()
+            .ok_or_else(|| Error::InvalidResponse("messages response is not an object".into()))?;
 
         let content = body
             .get("content")
             .and_then(Value::as_array)
-            .ok_or(CoreError::MissingField("content"))?;
+            .ok_or(Error::MissingField("content"))?;
         // The route declines tool and thinking requests, so a non-text block
         // means the response carries something this path never asked for.
         // Decline rather than silently dropping it; the host falls back.
@@ -163,7 +164,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
             .iter()
             .any(|block| block.get("type").and_then(Value::as_str) != Some("text"))
         {
-            return Err(CoreError::Unsupported("non-text response content block"));
+            return Err(Error::Unsupported("non-text response content block"));
         }
         let text: String = content
             .iter()
@@ -173,7 +174,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         let usage = body
             .get("usage")
             .and_then(Value::as_object)
-            .ok_or(CoreError::MissingField("usage"))?;
+            .ok_or(Error::MissingField("usage"))?;
         let field = |name: &str| usage.get(name).and_then(Value::as_u64).unwrap_or(0);
 
         Ok(ChatCompletionsResponse {
@@ -181,7 +182,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
             model: body
                 .get("model")
                 .and_then(Value::as_str)
-                .ok_or(CoreError::MissingField("model"))?
+                .ok_or(Error::MissingField("model"))?
                 .to_string(),
             choices: vec![ChatCompletionsChoice {
                 index: 0,

litellm-rust/crates/core/src/providers/anthropic/messages/transformation.rs

@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::messages::transformation::{AnthropicMessagesProviderConfig, MessagesAuthStrategy};
 
 const ANTHROPIC_API_KEY_ENV: &str = "ANTHROPIC_API_KEY";
@@ -17,12 +17,12 @@ pub fn non_empty(value: Option<&str>) -> Option<&str> {
 pub fn resolve_anthropic_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     non_empty(api_key)
         .map(str::to_string)
         .or_else(|| env_lookup(ANTHROPIC_API_KEY_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Anthropic API Key - Set `api_key` or the ANTHROPIC_API_KEY \
                  environment variable"
                     .to_string(),
@@ -52,7 +52,7 @@ impl AnthropicMessagesProviderConfig for AnthropicMessagesConfig {
         api_base: Option<&str>,
         _model: &str,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         Ok(complete_anthropic_url(api_base, env_lookup))
     }
 
@@ -60,7 +60,7 @@ impl AnthropicMessagesProviderConfig for AnthropicMessagesConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_anthropic_api_key(api_key, env_lookup)
     }
 
@@ -121,7 +121,7 @@ mod tests {
         );
         assert!(matches!(
             resolve_anthropic_api_key(None, &|_| None).expect_err("missing key"),
-            CoreError::Auth(_)
+            Error::Auth(_)
         ));
     }
 

litellm-rust/crates/core/src/providers/azure_ai/messages/transformation.rs

@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::messages::transformation::{AnthropicMessagesProviderConfig, MessagesAuthStrategy};
 use crate::messages::types::{
     AnthropicMessage, AnthropicMessagesRequest, AnthropicMessagesResponse, ContentBlock,
@@ -28,12 +28,12 @@ pub const AZURE_ANTHROPIC_MESSAGES_CONFIG: AzureAnthropicMessagesConfig =
 pub fn resolve_azure_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     non_empty(api_key)
         .map(str::to_string)
         .or_else(|| env_lookup(AZURE_API_KEY_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Azure API Key - Set `api_key` or the AZURE_API_KEY environment variable"
                     .to_string(),
             )
@@ -43,12 +43,12 @@ pub fn resolve_azure_api_key(
 pub fn complete_azure_anthropic_url(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let api_base = non_empty(api_base)
         .map(str::to_string)
         .or_else(|| env_lookup(AZURE_API_BASE_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Azure API Base - Set `api_base` or the AZURE_API_BASE environment variable. \
                  Expected format: https://<resource-name>.services.ai.azure.com/anthropic"
                     .to_string(),
@@ -147,7 +147,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
         api_base: Option<&str>,
         _model: &str,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_azure_anthropic_url(api_base, env_lookup)
     }
 
@@ -155,7 +155,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_azure_api_key(api_key, env_lookup)
     }
 
@@ -174,7 +174,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
     fn transform_request(
         &self,
         request: AnthropicMessagesRequest,
-    ) -> CoreResult<AnthropicMessagesRequest> {
+    ) -> Result<AnthropicMessagesRequest, Error> {
         let mut request = fold_system_role_messages(request);
         if let Some(system) = request.system.as_mut() {
             strip_scope_from_system(system);
@@ -190,7 +190,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
         &self,
         model: &str,
         response: AnthropicMessagesResponse,
-    ) -> CoreResult<AnthropicMessagesResponse> {
+    ) -> Result<AnthropicMessagesResponse, Error> {
         self.anthropic.transform_response(model, response)
     }
 }
@@ -268,7 +268,7 @@ mod tests {
             "https://env.services.ai.azure.com/anthropic/v1/messages"
         );
         let err = complete_azure_anthropic_url(Some("  "), &|_| None).expect_err("missing base");
-        assert!(matches!(err, CoreError::Auth(_)));
+        assert!(matches!(err, Error::Auth(_)));
     }
 
     #[test]
@@ -284,7 +284,7 @@ mod tests {
         );
         assert!(matches!(
             resolve_azure_api_key(None, &|_| None).expect_err("missing key"),
-            CoreError::Auth(_)
+            Error::Auth(_)
         ));
     }
 

litellm-rust/crates/core/src/providers/azure_ai/ocr/transformation.rs

@@ -1,6 +1,6 @@
 use std::collections::BTreeSet;
 
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 use crate::ocr::transformation::{OcrAuthStrategy, OcrProviderConfig, OcrResponseHandling};
 use crate::ocr::types::{OcrRequestData, OcrResponseData};
 use serde_json::{Map, Value, json};
@@ -32,17 +32,17 @@ fn resolve_value(
     env_name: &str,
     env_lookup: &dyn Fn(&str) -> Option<String>,
     missing_message: &str,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     non_empty(explicit)
         .map(str::to_string)
         .or_else(|| env_lookup(env_name).filter(|value| !value.trim().is_empty()))
-        .ok_or_else(|| CoreError::Auth(missing_message.to_string()))
+        .ok_or_else(|| Error::Auth(missing_message.to_string()))
 }
 
 pub fn resolve_azure_ai_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_key,
         AZURE_AI_API_KEY_ENV,
@@ -54,7 +54,7 @@ pub fn resolve_azure_ai_api_key(
 pub fn resolve_azure_ai_api_base(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_base,
         AZURE_AI_API_BASE_ENV,
@@ -66,7 +66,7 @@ pub fn resolve_azure_ai_api_base(
 pub fn complete_azure_ai_url(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let base = resolve_azure_ai_api_base(api_base, env_lookup)?;
     Ok(format!(
         "{}/providers/mistral/azure/ocr",
@@ -77,7 +77,7 @@ pub fn complete_azure_ai_url(
 pub fn resolve_document_intelligence_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_key,
         AZURE_DOCUMENT_INTELLIGENCE_API_KEY_ENV,
@@ -89,7 +89,7 @@ pub fn resolve_document_intelligence_api_key(
 pub fn resolve_document_intelligence_endpoint(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_base,
         AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT_ENV,
@@ -127,7 +127,7 @@ fn pages_token_is_valid(token: &str) -> bool {
     }
 }
 
-fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
+fn normalize_pages_param(pages: &Value) -> Result<Option<String>, Error> {
     match pages {
         Value::String(value) => {
             let normalized = value
@@ -138,7 +138,7 @@ fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
             if normalized.split(',').all(pages_token_is_valid) {
                 Ok(Some(normalized))
             } else {
-                Err(CoreError::InvalidRequest(format!(
+                Err(Error::InvalidRequest(format!(
                     "Invalid `pages` string for Azure Document Intelligence: {value:?}. Expected format like '1-3,5,7-9'."
                 )))
             }
@@ -152,7 +152,7 @@ fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
                 for value in values {
                     let page = value.as_i64().expect("checked is_i64");
                     if page < 0 {
-                        return Err(CoreError::InvalidRequest(
+                        return Err(Error::InvalidRequest(
                             "`pages` integers must be >= 0 (Mistral 0-based indices)".to_string(),
                         ));
                     }
@@ -176,16 +176,16 @@ fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
                 if normalized.split(',').all(pages_token_is_valid) {
                     return Ok(Some(normalized));
                 }
-                return Err(CoreError::InvalidRequest(format!(
+                return Err(Error::InvalidRequest(format!(
                     "Invalid `pages` list for Azure Document Intelligence: {values:?}. Expected tokens like '1' or '3-5'."
                 )));
             }
-            Err(CoreError::InvalidRequest(
+            Err(Error::InvalidRequest(
                 "`pages` must be a list[int] (0-based, Mistral-style) or a string like '1-3,5,7-9'."
                     .to_string(),
             ))
         }
-        _ => Err(CoreError::InvalidRequest(
+        _ => Err(Error::InvalidRequest(
             "`pages` must be a list[int] (0-based, Mistral-style) or a string like '1-3,5,7-9'."
                 .to_string(),
         )),
@@ -197,7 +197,7 @@ pub fn complete_document_intelligence_url(
     model: &str,
     optional_params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let endpoint = resolve_document_intelligence_endpoint(api_base, env_lookup)?;
     let mut url = format!(
         "{}/documentintelligence/documentModels/{}:analyze?api-version={}",
@@ -216,20 +216,20 @@ pub fn complete_document_intelligence_url(
     Ok(url)
 }
 
-fn document_url_from_mistral_document(document: &Value) -> CoreResult<&str> {
-    let object = document.as_object().ok_or_else(|| CoreError::InvalidType {
+fn document_url_from_mistral_document(document: &Value) -> Result<&str, Error> {
+    let object = document.as_object().ok_or_else(|| Error::InvalidType {
         expected: "object",
         actual: json_type_name(document),
     })?;
     let doc_type = object
         .get("type")
         .and_then(Value::as_str)
-        .ok_or(CoreError::MissingField("document.type"))?;
+        .ok_or(Error::MissingField("document.type"))?;
     let field_name = match doc_type {
         "document_url" => "document_url",
         "image_url" => "image_url",
         other => {
-            return Err(CoreError::InvalidRequest(format!(
+            return Err(Error::InvalidRequest(format!(
                 "Invalid document type: {other}. Must be 'document_url' or 'image_url'"
             )));
         }
@@ -238,7 +238,7 @@ fn document_url_from_mistral_document(document: &Value) -> CoreResult<&str> {
         .get(field_name)
         .and_then(Value::as_str)
         .filter(|value| !value.is_empty())
-        .ok_or(CoreError::MissingField(field_name))
+        .ok_or(Error::MissingField(field_name))
 }
 
 fn extract_base64_from_data_uri(data_uri: &str) -> &str {
@@ -290,7 +290,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_request(model, document, optional_params)
     }
 
@@ -298,7 +298,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_response(model, response_json)
     }
 
@@ -308,7 +308,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_azure_ai_url(api_base, env_lookup)
     }
 
@@ -316,7 +316,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_azure_ai_api_key(api_key, env_lookup)
     }
 
@@ -335,7 +335,7 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         _model: &str,
         document: Value,
         _optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         let document_url = document_url_from_mistral_document(&document)?;
         let mut data = Map::new();
         if document_url.starts_with("data:") {
@@ -359,19 +359,19 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         let response = response_json
             .as_object()
-            .ok_or_else(|| CoreError::InvalidType {
+            .ok_or_else(|| Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&response_json),
             })?;
         let status = response
             .get("status")
             .and_then(Value::as_str)
-            .ok_or(CoreError::MissingField("status"))?;
+            .ok_or(Error::MissingField("status"))?;
         if status != "succeeded" {
-            return Err(CoreError::InvalidResponse(format!(
+            return Err(Error::InvalidResponse(format!(
                 "Azure Document Intelligence analysis failed with status: {status}"
             )));
         }
@@ -414,7 +414,7 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_document_intelligence_url(api_base, model, optional_params, env_lookup)
     }
 
@@ -422,7 +422,7 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_document_intelligence_api_key(api_key, env_lookup)
     }
 

litellm-rust/crates/core/src/providers/bedrock/audio_transcription.rs

@@ -6,7 +6,7 @@ use crate::audio_transcription::transformation::{
 use crate::audio_transcription::types::{
     AudioTranscriptionRequestData, AudioTranscriptionResponseData,
 };
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 
 pub use super::aws_base::{aws_auth_config, bedrock_model_id_and_region, resolve_bedrock_region};
 use super::constants::{BEDROCK_RUNTIME_ENDPOINT_TEMPLATE, BEDROCK_SERVICE};
@@ -18,8 +18,8 @@ pub static BEDROCK_AUDIO_TRANSCRIPTION_CONFIG: BedrockAudioTranscriptionConfig =
 
 pub struct BedrockAudioTranscriptionConfig;
 
-fn audio_fields(audio: Value) -> CoreResult<(String, String)> {
-    let object = audio.as_object().ok_or_else(|| CoreError::InvalidType {
+fn audio_fields(audio: Value) -> Result<(String, String), Error> {
+    let object = audio.as_object().ok_or_else(|| Error::InvalidType {
         expected: "object",
         actual: json_type_name(&audio),
     })?;
@@ -27,13 +27,13 @@ fn audio_fields(audio: Value) -> CoreResult<(String, String)> {
         .get("data")
         .and_then(Value::as_str)
         .filter(|value| !value.is_empty())
-        .ok_or(CoreError::MissingField("audio.data"))?;
+        .ok_or(Error::MissingField("audio.data"))?;
     let format = object
         .get("format")
         .and_then(Value::as_str)
         .filter(|value| matches!(*value, "wav" | "mp3" | "flac" | "ogg"))
         .ok_or_else(|| {
-            CoreError::InvalidRequest("audio.format must be wav, mp3, flac, or ogg".to_string())
+            Error::InvalidRequest("audio.format must be wav, mp3, flac, or ogg".to_string())
         })?;
     Ok((data.to_string(), format.to_string()))
 }
@@ -55,7 +55,7 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         _model: &str,
         audio: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<AudioTranscriptionRequestData> {
+    ) -> Result<AudioTranscriptionRequestData, Error> {
         let (data, format) = audio_fields(audio)?;
         let mut instruction = "Transcribe the audio. Respond with only the transcript.".to_string();
         if let Some(language) = optional_string(&optional_params, "language") {
@@ -87,14 +87,14 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         &self,
         _model: &str,
         response_json: Value,
-    ) -> CoreResult<AudioTranscriptionResponseData> {
+    ) -> Result<AudioTranscriptionResponseData, Error> {
         let content = response_json
             .get("output")
             .and_then(|value| value.get("message"))
             .and_then(|value| value.get("content"))
             .and_then(Value::as_array)
             .ok_or_else(|| {
-                CoreError::InvalidResponse("Bedrock response has no output content".to_string())
+                Error::InvalidResponse("Bedrock response has no output content".to_string())
             })?;
         let mut text = String::new();
         for block in content {
@@ -111,7 +111,7 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         let (model_id, model_region) = bedrock_model_id_and_region(model);
         let region = resolve_bedrock_region(model_region.as_deref(), optional_params, env_lookup);
         let endpoint = optional_params
@@ -133,7 +133,7 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<AudioTranscriptionAuth> {
+    ) -> Result<AudioTranscriptionAuth, Error> {
         let (_, model_region) = bedrock_model_id_and_region(model);
         Ok(AudioTranscriptionAuth::AwsSigV4 {
             region: resolve_bedrock_region(model_region.as_deref(), optional_params, env_lookup),

litellm-rust/crates/core/src/providers/bedrock/aws_base.rs

@@ -4,7 +4,7 @@ use std::time::Duration;
 use std::time::{SystemTime, UNIX_EPOCH};
 
 use crate::caching::in_memory_cache::InMemoryCache;
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use aws_credential_types::Credentials;
 use aws_credential_types::provider::ProvideCredentials;
 use aws_sigv4::http_request::{
@@ -197,7 +197,7 @@ pub fn classify_auth(
 pub async fn resolve_credentials(
     config: AwsAuthConfig,
     env_lookup: &(dyn Fn(&str) -> Option<String> + Sync),
-) -> CoreResult<Credentials> {
+) -> Result<Credentials, Error> {
     let resolved = config.clone().with_environment(env_lookup);
     let flow = classify_auth(config, env_lookup);
     match flow {
@@ -244,9 +244,10 @@ pub async fn resolve_credentials(
             let provider = aws_config::profile::ProfileFileCredentialsProvider::builder()
                 .profile_name(name)
                 .build();
-            provider.provide_credentials().await.map_err(|error| {
-                CoreError::Auth(format!("AWS profile credentials failed: {error}"))
-            })
+            provider
+                .provide_credentials()
+                .await
+                .map_err(|error| Error::Auth(format!("AWS profile credentials failed: {error}")))
         }
         AwsAuthFlow::AssumeRole { role, session_name } => {
             if is_already_running_as_role(&role, &resolved).await? {
@@ -260,7 +261,7 @@ pub async fn resolve_credentials(
                         .build()
                         .await;
                 let credentials = provider.provide_credentials().await.map_err(|error| {
-                    CoreError::Auth(format!("AWS default credentials failed: {error}"))
+                    Error::Auth(format!("AWS default credentials failed: {error}"))
                 })?;
                 set_cached_credentials(
                     key,
@@ -301,7 +302,7 @@ pub async fn resolve_credentials(
             provider
                 .provide_credentials()
                 .await
-                .map_err(|error| CoreError::Auth(format!("AWS role credentials failed: {error}")))
+                .map_err(|error| Error::Auth(format!("AWS role credentials failed: {error}")))
         }
         AwsAuthFlow::WebIdentity {
             token,
@@ -325,13 +326,13 @@ pub async fn resolve_credentials(
                 .send()
                 .await
                 .map_err(|error| {
-                    CoreError::Auth(format!("AWS web identity credentials failed: {error}"))
+                    Error::Auth(format!("AWS web identity credentials failed: {error}"))
                 })?;
             let credentials = response.credentials().ok_or_else(|| {
-                CoreError::Auth("AWS web identity response had no credentials".to_string())
+                Error::Auth("AWS web identity response had no credentials".to_string())
             })?;
             let expiration = SystemTime::try_from(*credentials.expiration()).map_err(|error| {
-                CoreError::Auth(format!("AWS web identity expiration was invalid: {error}"))
+                Error::Auth(format!("AWS web identity expiration was invalid: {error}"))
             })?;
             Ok(Credentials::new(
                 credentials.access_key_id(),
@@ -350,9 +351,10 @@ pub async fn resolve_credentials(
                 aws_config::default_provider::credentials::DefaultCredentialsChain::builder()
                     .build()
                     .await;
-            let credentials = provider.provide_credentials().await.map_err(|error| {
-                CoreError::Auth(format!("AWS default credentials failed: {error}"))
-            })?;
+            let credentials = provider
+                .provide_credentials()
+                .await
+                .map_err(|error| Error::Auth(format!("AWS default credentials failed: {error}")))?;
             set_cached_credentials(
                 key,
                 credentials.clone(),
@@ -363,7 +365,7 @@ pub async fn resolve_credentials(
     }
 }
 
-async fn is_already_running_as_role(role: &str, config: &AwsAuthConfig) -> CoreResult<bool> {
+async fn is_already_running_as_role(role: &str, config: &AwsAuthConfig) -> Result<bool, Error> {
     if role_identity(role).is_none() {
         return Ok(false);
     }
@@ -437,7 +439,7 @@ pub fn sign_bedrock_post(
     region: &str,
     credentials: &Credentials,
     signing_time: SystemTime,
-) -> CoreResult<BTreeMap<String, String>> {
+) -> Result<BTreeMap<String, String>, Error> {
     let identity: Identity = credentials.clone().into();
     let params = v4::SigningParams::builder()
         .identity(&identity)
@@ -447,14 +449,14 @@ pub fn sign_bedrock_post(
         .settings(SigningSettings::default())
         .build()
         .map(SigningParams::from)
-        .map_err(|error| CoreError::Auth(format!("AWS signing parameters failed: {error}")))?;
+        .map_err(|error| Error::Auth(format!("AWS signing parameters failed: {error}")))?;
     let header_refs = headers
         .iter()
         .map(|(name, value)| (name.as_str(), value.as_str()));
     let request = SignableRequest::new("POST", url, header_refs, SignableBody::Bytes(body))
-        .map_err(|error| CoreError::Auth(format!("AWS signable request failed: {error}")))?;
+        .map_err(|error| Error::Auth(format!("AWS signable request failed: {error}")))?;
     let (instructions, _) = sign(request, &params)
-        .map_err(|error| CoreError::Auth(format!("AWS request signing failed: {error}")))?
+        .map_err(|error| Error::Auth(format!("AWS request signing failed: {error}")))?
         .into_parts();
     Ok(instructions
         .headers()

litellm-rust/crates/core/src/providers/bedrock/chat_completions/tests.rs

@@ -1,4 +1,5 @@
 use super::*;
+use crate::Error;
 use serde_json::json;
 
 fn messages(value: Value) -> Vec<ChatMessage> {
@@ -23,7 +24,7 @@ fn transform(msgs: Value, opts: Value) -> Value {
         .body
 }
 
-fn transform_response(body: Value) -> CoreResult<ChatCompletionsResponse> {
+fn transform_response(body: Value) -> Result<ChatCompletionsResponse, Error> {
     BEDROCK_CHAT_COMPLETIONS_CONFIG.transform_response(
         "anthropic.claude-sonnet-4-5-v1:0",
         ProviderChatResponseData { body },
@@ -478,25 +479,22 @@ fn declines_a_response_carrying_a_tool_use_block() {
         "usage": {"inputTokens": 1, "outputTokens": 1}
     }))
     .expect_err("tool use block");
-    assert_eq!(
-        err,
-        CoreError::Unsupported("non-text response content block")
-    );
+    assert_eq!(err, Error::Unsupported("non-text response content block"));
 }
 
 #[test]
 fn errors_on_a_response_missing_required_fields() {
     assert_eq!(
         transform_response(json!("nope")).expect_err("not an object"),
-        CoreError::InvalidResponse("converse response is not an object".to_string())
+        Error::InvalidResponse("converse response is not an object".to_string())
     );
     assert_eq!(
         transform_response(json!({"usage": {}})).expect_err("no output"),
-        CoreError::MissingField("output.message.content")
+        Error::MissingField("output.message.content")
     );
     assert_eq!(
         transform_response(json!({"output": {"message": {"content": []}}})).expect_err("no usage"),
-        CoreError::MissingField("usage")
+        Error::MissingField("usage")
     );
 }
 

litellm-rust/crates/core/src/providers/bedrock/chat_completions/transformation.rs

@@ -11,7 +11,7 @@ use crate::chat_completions::types::{
     ChatCompletionsUsage, ChatMessage, ChatMessageContent, ProviderChatRequestData,
     ProviderChatResponseData,
 };
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 
 use super::super::aws_base::{bedrock_model_id_and_region, resolve_bedrock_region};
 use super::super::constants::{AWS_BEARER_TOKEN_BEDROCK, BEDROCK_RUNTIME_ENDPOINT_TEMPLATE};
@@ -110,7 +110,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         let (model_id, model_region) = bedrock_model_id_and_region(model);
         let region = resolve_bedrock_region(model_region.as_deref(), optional_params, env_lookup);
         let endpoint = optional_params
@@ -137,7 +137,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<ChatCompletionsAuth> {
+    ) -> Result<ChatCompletionsAuth, Error> {
         // Python reads `api_key` as the Bedrock bearer token and consults the
         // env only when the caller passed none, so a caller-supplied empty key
         // falls through to SigV4 without reaching for the environment. An
@@ -208,7 +208,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         _model: &str,
         messages: Vec<ChatMessage>,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<ProviderChatRequestData> {
+    ) -> Result<ProviderChatRequestData, Error> {
         Ok(ProviderChatRequestData {
             body: converse_body(&build_conversation(&messages), &optional_params),
         })
@@ -218,17 +218,18 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         &self,
         model: &str,
         response: ProviderChatResponseData,
-    ) -> CoreResult<ChatCompletionsResponse> {
-        let body = response.body.as_object().ok_or_else(|| {
-            CoreError::InvalidResponse("converse response is not an object".into())
-        })?;
+    ) -> Result<ChatCompletionsResponse, Error> {
+        let body = response
+            .body
+            .as_object()
+            .ok_or_else(|| Error::InvalidResponse("converse response is not an object".into()))?;
 
         let content = body
             .get("output")
             .and_then(|output| output.get("message"))
             .and_then(|message| message.get("content"))
             .and_then(Value::as_array)
-            .ok_or(CoreError::MissingField("output.message.content"))?;
+            .ok_or(Error::MissingField("output.message.content"))?;
         // The route declines tool requests, so anything other than a text block
         // is something this path never asked for. Decline; the host falls back.
         if content.iter().any(|block| {
@@ -236,7 +237,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
                 .as_object()
                 .is_none_or(|block| block.len() != 1 || !block.contains_key("text"))
         }) {
-            return Err(CoreError::Unsupported("non-text response content block"));
+            return Err(Error::Unsupported("non-text response content block"));
         }
         let text: String = content
             .iter()
@@ -246,7 +247,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         let usage = body
             .get("usage")
             .and_then(Value::as_object)
-            .ok_or(CoreError::MissingField("usage"))?;
+            .ok_or(Error::MissingField("usage"))?;
         let field = |name: &str| usage.get(name).and_then(Value::as_u64).unwrap_or(0);
         let computed = usage_from_parts(
             field("inputTokens"),

litellm-rust/crates/core/src/providers/mistral/ocr/transformation.rs

@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 use crate::ocr::transformation::OcrProviderConfig;
 use crate::ocr::types::{OcrRequestData, OcrResponseData};
 use serde_json::{Map, Value};
@@ -47,7 +47,7 @@ pub fn complete_url(api_base: Option<&str>) -> String {
 
 /// Resolve the Mistral API key from the explicit param or the environment.
 ///
-/// Blank/whitespace values are treated as absent. Returns `CoreError::Auth`
+/// Blank/whitespace values are treated as absent. Returns `Error::Auth`
 /// when no usable key is available.
 ///
 /// Note: the env fallback only reads the process environment. Secret-manager
@@ -56,13 +56,13 @@ pub fn complete_url(api_base: Option<&str>) -> String {
 pub fn resolve_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|key| !key.is_empty())
         .map(str::to_string)
         .or_else(|| env_lookup(MISTRAL_API_KEY_ENV).filter(|key| !key.trim().is_empty()))
-        .ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
+        .ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
 }
 
 pub struct MistralOcrConfig;
@@ -79,9 +79,9 @@ impl OcrProviderConfig for MistralOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         if !document.is_object() {
-            return Err(CoreError::InvalidType {
+            return Err(Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&document),
             });
@@ -104,10 +104,10 @@ impl OcrProviderConfig for MistralOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         let response_object = response_json
             .as_object()
-            .ok_or_else(|| CoreError::InvalidType {
+            .ok_or_else(|| Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&response_json),
             })?;
@@ -140,7 +140,7 @@ impl OcrProviderConfig for MistralOcrConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         _env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         Ok(complete_url(api_base))
     }
 
@@ -148,7 +148,7 @@ impl OcrProviderConfig for MistralOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_api_key(api_key, env_lookup)
     }
 }
@@ -165,11 +165,11 @@ pub fn transform_ocr_request(
     model: &str,
     document: Value,
     optional_params: Map<String, Value>,
-) -> CoreResult<OcrRequestData> {
+) -> Result<OcrRequestData, Error> {
     MISTRAL_OCR_CONFIG.transform_ocr_request(model, document, optional_params)
 }
 
-pub fn transform_ocr_response(model: &str, response_json: Value) -> CoreResult<OcrResponseData> {
+pub fn transform_ocr_response(model: &str, response_json: Value) -> Result<OcrResponseData, Error> {
     MISTRAL_OCR_CONFIG.transform_ocr_response(model, response_json)
 }
 
@@ -250,7 +250,7 @@ mod tests {
 
         assert_eq!(
             err,
-            CoreError::InvalidType {
+            Error::InvalidType {
                 expected: "object",
                 actual: "string",
             }
@@ -307,6 +307,6 @@ mod tests {
     #[test]
     fn resolve_api_key_errors_when_absent() {
         let err = resolve_api_key(None, &|_| None).expect_err("missing key should error");
-        assert_eq!(err, CoreError::Auth(MISSING_KEY_MESSAGE.to_string()));
+        assert_eq!(err, Error::Auth(MISSING_KEY_MESSAGE.to_string()));
     }
 }

litellm-rust/crates/core/src/providers/openai/realtime/transformation.rs

@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::realtime::transformation::RealtimeProviderConfig;
 use crate::realtime::types::{RealtimeEvent, RealtimeTransformResult};
 
@@ -72,7 +72,7 @@ impl RealtimeProviderConfig for OpenAiRealtimeConfig {
         &self,
         event: &RealtimeEvent,
         _model: &str,
-    ) -> CoreResult<RealtimeTransformResult> {
+    ) -> Result<RealtimeTransformResult, Error> {
         Ok(RealtimeTransformResult::passthrough(event.clone()))
     }
 
@@ -80,7 +80,7 @@ impl RealtimeProviderConfig for OpenAiRealtimeConfig {
         &self,
         event: &RealtimeEvent,
         _model: &str,
-    ) -> CoreResult<RealtimeTransformResult> {
+    ) -> Result<RealtimeTransformResult, Error> {
         Ok(RealtimeTransformResult::passthrough(event.clone()))
     }
 }
@@ -88,14 +88,14 @@ impl RealtimeProviderConfig for OpenAiRealtimeConfig {
 pub fn transform_realtime_request(
     event: &RealtimeEvent,
     model: &str,
-) -> CoreResult<RealtimeTransformResult> {
+) -> Result<RealtimeTransformResult, Error> {
     OPENAI_REALTIME_CONFIG.transform_realtime_request(event, model)
 }
 
 pub fn transform_realtime_response(
     event: &RealtimeEvent,
     model: &str,
-) -> CoreResult<RealtimeTransformResult> {
+) -> Result<RealtimeTransformResult, Error> {
     OPENAI_REALTIME_CONFIG.transform_realtime_response(event, model)
 }
 

litellm-rust/crates/core/src/providers/openai/responses/transformation.rs

@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::responses::types::{ResponsesWsEvent, ResponsesWsTransformResult};
 use crate::responses::websocket::{ResponsesWebSocketProviderConfig, enforce_model};
 
@@ -15,7 +15,7 @@ impl ResponsesWebSocketProviderConfig for OpenAIResponsesWsConfig {
         &self,
         event: &ResponsesWsEvent,
         model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult> {
+    ) -> Result<ResponsesWsTransformResult, Error> {
         Ok(ResponsesWsTransformResult::passthrough(enforce_model(
             event, model,
         )))
@@ -25,7 +25,7 @@ impl ResponsesWebSocketProviderConfig for OpenAIResponsesWsConfig {
         &self,
         event: &ResponsesWsEvent,
         _model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult> {
+    ) -> Result<ResponsesWsTransformResult, Error> {
         Ok(ResponsesWsTransformResult::passthrough(event.clone()))
     }
 }

litellm-rust/crates/core/src/providers/vertex_ai/ocr/transformation.rs

@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 use crate::ocr::transformation::OcrProviderConfig;
 use crate::ocr::types::{OcrRequestData, OcrResponseData};
 use serde_json::{Map, Value, json};
@@ -43,7 +43,7 @@ pub fn is_deepseek_model(model: &str) -> bool {
 pub fn resolve_vertex_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|key| !key.is_empty())
@@ -51,7 +51,7 @@ pub fn resolve_vertex_api_key(
         .or_else(|| env_lookup(VERTEX_AI_API_KEY_ENV).filter(|key| !key.trim().is_empty()))
         .or_else(|| env_lookup(VERTEXAI_API_KEY_ENV).filter(|key| !key.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Vertex AI access token - pass api_key or provide Authorization via extra_headers"
                     .to_string(),
             )
@@ -61,12 +61,12 @@ pub fn resolve_vertex_api_key(
 fn vertex_project(
     params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     string_param(params, &["vertex_project", "vertex_ai_project"])
         .map(str::to_string)
         .or_else(|| env_lookup(VERTEXAI_PROJECT_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::InvalidRequest(
+            Error::InvalidRequest(
                 "Missing vertex_project - Set VERTEXAI_PROJECT environment variable or pass vertex_project parameter"
                     .to_string(),
             )
@@ -99,7 +99,7 @@ pub fn complete_vertex_mistral_url(
     model: &str,
     optional_params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let project = vertex_project(optional_params, env_lookup)?;
     let location = vertex_location(optional_params, env_lookup);
     let base = vertex_mistral_api_base(api_base, &location);
@@ -112,7 +112,7 @@ pub fn complete_vertex_deepseek_url(
     api_base: Option<&str>,
     optional_params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let project = vertex_project(optional_params, env_lookup)?;
     let location = vertex_location(optional_params, env_lookup);
     let base = api_base
@@ -125,20 +125,20 @@ pub fn complete_vertex_deepseek_url(
     ))
 }
 
-fn document_content_item(document: &Value) -> CoreResult<Value> {
-    let object = document.as_object().ok_or_else(|| CoreError::InvalidType {
+fn document_content_item(document: &Value) -> Result<Value, Error> {
+    let object = document.as_object().ok_or_else(|| Error::InvalidType {
         expected: "object",
         actual: json_type_name(document),
     })?;
     let doc_type = object
         .get("type")
         .and_then(Value::as_str)
-        .ok_or(CoreError::MissingField("document.type"))?;
+        .ok_or(Error::MissingField("document.type"))?;
     let url_field = match doc_type {
         "image_url" => "image_url",
         "document_url" => "document_url",
         other => {
-            return Err(CoreError::InvalidRequest(format!(
+            return Err(Error::InvalidRequest(format!(
                 "Unsupported document type: {other}. Expected 'image_url' or 'document_url'"
             )));
         }
@@ -147,7 +147,7 @@ fn document_content_item(document: &Value) -> CoreResult<Value> {
         .get(url_field)
         .and_then(Value::as_str)
         .filter(|value| !value.is_empty())
-        .ok_or(CoreError::MissingField(url_field))?;
+        .ok_or(Error::MissingField(url_field))?;
 
     Ok(json!({
         "type": "image_url",
@@ -163,7 +163,7 @@ fn deepseek_model_name(model: &str) -> String {
     }
 }
 
-fn first_choice_content(response: &Value) -> CoreResult<Value> {
+fn first_choice_content(response: &Value) -> Result<Value, Error> {
     response
         .get("choices")
         .and_then(Value::as_array)
@@ -176,9 +176,7 @@ fn first_choice_content(response: &Value) -> CoreResult<Value> {
             Value::Object(_) => true,
             _ => false,
         })
-        .ok_or_else(|| {
-            CoreError::InvalidResponse("No content in DeepSeek OCR response".to_string())
-        })
+        .ok_or_else(|| Error::InvalidResponse("No content in DeepSeek OCR response".to_string()))
 }
 
 fn ocr_data_from_content(content: Value, usage: Option<Value>, model: &str) -> Value {
@@ -219,7 +217,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_request(model, document, optional_params)
     }
 
@@ -227,7 +225,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_response(model, response_json)
     }
 
@@ -237,7 +235,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_vertex_mistral_url(api_base, model, optional_params, env_lookup)
     }
 
@@ -245,7 +243,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_vertex_api_key(api_key, env_lookup)
     }
 
@@ -264,7 +262,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         let mut data = Map::new();
         data.insert(
             "model".to_string(),
@@ -289,10 +287,10 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         let response = response_json
             .as_object()
-            .ok_or_else(|| CoreError::InvalidType {
+            .ok_or_else(|| Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&response_json),
             })?;
@@ -314,7 +312,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
             });
         }
 
-        let object = ocr_data.as_object().ok_or_else(|| CoreError::InvalidType {
+        let object = ocr_data.as_object().ok_or_else(|| Error::InvalidType {
             expected: "object",
             actual: json_type_name(&ocr_data),
         })?;
@@ -346,7 +344,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         _model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_vertex_deepseek_url(api_base, optional_params, env_lookup)
     }
 
@@ -354,7 +352,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_vertex_api_key(api_key, env_lookup)
     }
 }

litellm-rust/crates/core/src/realtime/transformation.rs

@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::realtime::types::{RealtimeEvent, RealtimeTransformResult};
 
 pub trait RealtimeProviderConfig {
@@ -11,12 +11,12 @@ pub trait RealtimeProviderConfig {
         &self,
         event: &RealtimeEvent,
         model: &str,
-    ) -> CoreResult<RealtimeTransformResult>;
+    ) -> Result<RealtimeTransformResult, Error>;
 
     /// Transform a backend → client event before it is forwarded downstream.
     fn transform_realtime_response(
         &self,
         event: &RealtimeEvent,
         model: &str,
-    ) -> CoreResult<RealtimeTransformResult>;
+    ) -> Result<RealtimeTransformResult, Error>;
 }

litellm-rust/crates/core/src/responses/instrumentation.rs

@@ -5,9 +5,9 @@ use std::time::{SystemTime, UNIX_EPOCH};
 
 use serde_json::Value;
 
+use crate::Error;
 use crate::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
 use crate::responses::types::{ResponsesWsEvent, ResponsesWsEventType};
-use crate::{CoreError, CoreResult};
 
 #[derive(Clone, Debug, Default, PartialEq, Eq)]
 pub struct ResponsesWsUsage {
@@ -205,7 +205,7 @@ impl ResponsesWsInstrumentation {
     }
 }
 
-type LifecycleFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
+type LifecycleFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
 
 impl CallLifecycleHooks<(), (), ()> for ResponsesWsInstrumentation {
     type PreCallFuture<'a> = LifecycleFuture<'a, ()>;
@@ -246,7 +246,7 @@ impl CallLifecycleHooks<(), (), ()> for ResponsesWsInstrumentation {
     fn async_log_failure_event<'a>(
         &'a self,
         _context: &'a CallLifecycleContext,
-        _error: &'a CoreError,
+        _error: &'a Error,
         _timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a> {
         Box::pin(async move {
@@ -342,7 +342,7 @@ mod tests {
                 ),
                 (),
                 &instrumentation,
-                |_| async { Ok::<(), CoreError>(()) },
+                |_| async { Ok::<(), Error>(()) },
             )
             .await;
 

litellm-rust/crates/core/src/responses/websocket.rs

@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::constants::{OPENAI_RESPONSES_DEFAULT_API_BASE, OPENAI_RESPONSES_PATH};
 use crate::responses::types::{ResponsesWsEvent, ResponsesWsEventType, ResponsesWsTransformResult};
 
@@ -19,13 +19,13 @@ pub trait ResponsesWebSocketProviderConfig: Sync {
         &self,
         event: &ResponsesWsEvent,
         model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult>;
+    ) -> Result<ResponsesWsTransformResult, Error>;
 
     fn transform_ws_response(
         &self,
         event: &ResponsesWsEvent,
         model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult>;
+    ) -> Result<ResponsesWsTransformResult, Error>;
 }
 
 pub fn complete_websocket_url(

litellm-rust/crates/core/tests/workspace_crate_allowlist.rs

@@ -1,7 +1,8 @@
-//! Enforcement: the litellm-rust workspace has exactly three crates.
+//! Enforcement: the litellm-rust workspace has exactly four crates.
 //!
-//! `core` (pure translation), `ai-gateway` (routes + all network I/O), and
-//! `python-bridge` (the PyO3 cdylib). Adding or removing a crate must be a
+//! `core` (the Rust SDK), `ai-gateway` (the HTTP/WebSocket host),
+//! `python-interop` (domain-neutral PyO3 primitives), and `python-bridge` (the
+//! PyO3 cdylib). Adding or removing a crate must be a
 //! deliberate act: this test fails until the allowlist here is updated, forcing
 //! whoever changes the crate set to justify the new crate per the rule that a
 //! crate is a layer needing independent compilation / its own deps / a separate
@@ -16,10 +17,15 @@ use std::path::{Path, PathBuf};
 
 /// The one true crate set. Update BOTH this and `litellm-rust/AGENTS.md` when the
 /// workspace legitimately gains or loses a crate.
-const EXPECTED_MEMBERS: &[&str] = &["crates/core", "crates/ai-gateway", "crates/python-bridge"];
+const EXPECTED_MEMBERS: &[&str] = &[
+    "crates/core",
+    "crates/ai-gateway",
+    "crates/python-interop",
+    "crates/python-bridge",
+];
 
 /// The crate subdirectory names that must exist under `crates/`.
-const EXPECTED_CRATE_DIRS: &[&str] = &["core", "ai-gateway", "python-bridge"];
+const EXPECTED_CRATE_DIRS: &[&str] = &["core", "ai-gateway", "python-interop", "python-bridge"];
 
 const MISMATCH: &str = "litellm-rust crate set changed — update this allowlist AND litellm-rust/AGENTS.md, and justify the crate per the rule (crate = layer needing independent compilation / its own deps / a separate artifact).";
 

litellm-rust/crates/python-bridge/AGENTS.md

@@ -1,3 +1,3 @@
-litellm-python-bridge is the PyO3 cdylib that exposes Rust to the litellm Python SDK — a thin adapter (Python objects → Rust calls → Python results) over the litellm-core route entrypoints (e.g. `litellm_core::messages::messages`).
+litellm-python-bridge is the PyO3 cdylib that exposes LiteLLM Rust APIs to the Python SDK. Keep API registration, domain dependency wiring, request assembly, and Python exception mapping here. Put domain-neutral Python/Serde conversion and GIL primitives in litellm-python-interop.
 
 Keep it thin: no business logic, no transforms, no I/O orchestration — just marshal in/out and call the core entrypoint.

litellm-rust/crates/python-bridge/CLAUDE.md

@@ -5,8 +5,9 @@ Rules for `litellm-rust/crates/python-bridge`.
 ## Responsibility
 
 `python-bridge` is the PyO3 boundary between Python LiteLLM and Rust transforms.
-Keep this crate thin. It adapts Python objects to Rust payloads and returns
-Python-compatible dictionaries.
+Keep this crate thin. It exposes LiteLLM Rust APIs, assembles domain requests,
+maps domain errors to Python exceptions, and delegates generic conversion and
+GIL handling to `litellm-python-interop`.
 
 ## Bridge Shape
 

litellm-rust/crates/python-bridge/Cargo.toml

@@ -13,14 +13,14 @@ crate-type = ["cdylib"]
 default = ["abi3"]
 abi3 = ["pyo3/abi3-py310"]
 extension-module = ["pyo3/extension-module"]
+panic-test = []
 
 [dependencies]
 litellm-core = { workspace = true, features = ["bedrock-auth"] }
 litellm-ai-gateway = { workspace = true, default-features = false }
+litellm-python-interop.workspace = true
 pyo3.workspace = true
 pyo3-async-runtimes.workspace = true
-pythonize.workspace = true
-serde.workspace = true
 serde_json.workspace = true
 tokio.workspace = true
 

litellm-rust/crates/python-bridge/benches/serialization.rs

@@ -2,6 +2,7 @@ use std::hint::black_box;
 use std::time::Duration;
 
 use criterion::{BenchmarkId, Criterion, criterion_group, criterion_main};
+use litellm_python_interop::{from_py, to_py};
 use pyo3::prelude::*;
 use pyo3::types::PyDict;
 use serde_json::{Value, json};
@@ -25,7 +26,7 @@ fn former_json_roundtrip_from_py(py: Python<'_>, value: &Bound<'_, PyAny>) -> Va
 }
 
 fn pythonize_from_py(value: &Bound<'_, PyAny>) -> Value {
-    pythonize::depythonize(value).expect("payload should depythonize")
+    from_py(value).expect("payload should depythonize")
 }
 
 fn former_json_roundtrip_to_py(py: Python<'_>, value: &Value) -> Py<PyAny> {
@@ -37,12 +38,10 @@ fn former_json_roundtrip_to_py(py: Python<'_>, value: &Value) -> Py<PyAny> {
 }
 
 fn pythonize_to_py(py: Python<'_>, value: &Value) -> Py<PyAny> {
-    pythonize::pythonize(py, value)
-        .expect("response should pythonize")
-        .unbind()
+    to_py(py, value).expect("response should pythonize")
 }
 
-fn serialization(c: &mut Criterion) {
+fn bridge_serialization(c: &mut Criterion) {
     Python::initialize();
     Python::attach(|py| {
         for &(label, payload_bytes) in PAYLOAD_SIZES {
@@ -98,6 +97,6 @@ criterion_group! {
         .sample_size(20)
         .warm_up_time(Duration::from_secs(1))
         .measurement_time(Duration::from_secs(4));
-    targets = serialization
+    targets = bridge_serialization
 }
 criterion_main!(benches);

litellm-rust/crates/python-bridge/src/gil.rs

@@ -1,32 +0,0 @@
-//! GIL accounting.
-//!
-//! A single chokepoint for releasing the GIL around blocking work. Every
-//! blocking call in the bridge goes through [`release_gil`] instead of calling
-//! `Python::detach` directly, so the release count stays accurate and we
-//! have one place to extend later (timing histograms, per-call labels, etc.).
-
-use std::sync::atomic::{AtomicU64, Ordering};
-
-use pyo3::prelude::*;
-
-/// Number of times the bridge has released the GIL since process start.
-static GIL_RELEASES: AtomicU64 = AtomicU64::new(0);
-
-/// Release the GIL around `f`, recording the release.
-///
-/// `f` must not touch any Python state — that is what makes releasing the GIL
-/// safe. Returning the value back to Python re-acquires the GIL at the call
-/// site, after `f` has finished.
-pub fn release_gil<T, F>(py: Python<'_>, f: F) -> T
-where
-    F: FnOnce() -> T + Send,
-    T: Send,
-{
-    GIL_RELEASES.fetch_add(1, Ordering::Relaxed);
-    py.detach(f)
-}
-
-/// Total GIL releases performed by the bridge so far.
-pub fn release_count() -> u64 {
-    GIL_RELEASES.load(Ordering::Relaxed)
-}

litellm-rust/crates/python-bridge/src/lib.rs

@@ -10,19 +10,15 @@ use litellm_core::chat_completions::types::{ChatCompletionsRequest, ChatCompleti
 use litellm_core::chat_completions::{
     chat_completions as run_chat_completions, chat_completions_decline_reason,
 };
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::messages::messages as run_messages;
 use litellm_core::messages::types::{AnthropicMessagesResponse, MessagesRequest};
+use litellm_python_interop::{from_py, release_count, release_gil, to_py};
 use pyo3::exceptions::{PyRuntimeError, PyValueError};
 use pyo3::prelude::*;
 use pyo3::types::{PyAny, PyDict};
 use serde_json::{Map, Value};
 
-mod gil;
-mod marshal;
-
-use marshal::{from_py, to_py};
-
 pyo3::create_exception!(
     _native,
     RustBridgeDeclined,
@@ -58,13 +54,13 @@ fn chat_completions_response_to_py(
     to_py(py, &response)
 }
 
-fn core_error_to_pyerr(err: CoreError) -> PyErr {
+fn core_error_to_pyerr(err: Error) -> PyErr {
     match err {
-        CoreError::Auth(message) => PyValueError::new_err(message),
-        CoreError::InvalidProvider(_)
-        | CoreError::InvalidRequest(_)
-        | CoreError::InvalidType { .. }
-        | CoreError::MissingField(_) => PyValueError::new_err(err.to_string()),
+        Error::Auth(message) => PyValueError::new_err(message),
+        Error::InvalidProvider(_)
+        | Error::InvalidRequest(_)
+        | Error::InvalidType { .. }
+        | Error::MissingField(_) => PyValueError::new_err(err.to_string()),
         other => PyRuntimeError::new_err(other.to_string()),
     }
 }
@@ -75,22 +71,22 @@ fn core_error_to_pyerr(err: CoreError) -> PyErr {
 /// Everything raised before the request goes out is safe for the host to retry
 /// on its own path; anything after it is not, because the provider has already
 /// done the work and billed for it.
-fn chat_completions_error_to_pyerr(err: CoreError) -> PyErr {
+fn chat_completions_error_to_pyerr(err: Error) -> PyErr {
     match err {
-        CoreError::Unsupported(_)
-        | CoreError::Auth(_)
-        | CoreError::InvalidProvider(_)
-        | CoreError::InvalidRequest(_)
-        | CoreError::InvalidType { .. }
-        | CoreError::MissingField(_)
-        | CoreError::Routing(_)
+        Error::Unsupported(_)
+        | Error::Auth(_)
+        | Error::InvalidProvider(_)
+        | Error::InvalidRequest(_)
+        | Error::InvalidType { .. }
+        | Error::MissingField(_)
+        | Error::Routing(_)
         // Nothing reached the provider, so serving it on Python cannot double
         // bill and is the only way the caller gets an answer at all.
-        | CoreError::Connect(_) => RustBridgeDeclined::new_err(err.to_string()),
-        CoreError::Http { status, body } => {
+        | Error::Connect(_) => RustBridgeDeclined::new_err(err.to_string()),
+        Error::Http { status, body } => {
             RustUpstreamError::new_err((status, format!("{status}: {body}")))
         }
-        CoreError::Network(message) | CoreError::InvalidResponse(message) => {
+        Error::Network(message) | Error::InvalidResponse(message) => {
             RustUpstreamError::new_err((0u16, message))
         }
     }
@@ -230,7 +226,7 @@ fn ocr(
         timeout_seconds,
     )?;
 
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_ocr(OcrRequest {
             model: &model,
             document,
@@ -318,7 +314,7 @@ fn transcription(
     };
     let optional_params = optional_object_to_map(py, "optional_params", optional_params)?;
     let timeout = optional_timeout(timeout_seconds);
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_audio_transcription(
             AudioTranscriptionRequest {
                 model: &model,
@@ -419,7 +415,7 @@ fn messages(
     let (body, extra_headers, timeout) =
         marshal_messages_inputs(py, body, extra_headers, timeout_seconds)?;
 
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_messages(MessagesRequest {
             model: &model,
             body,
@@ -546,7 +542,7 @@ fn chat_completions(
         timeout_seconds,
     )?;
 
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_chat_completions(
             ChatCompletionsRequest {
                 model: &model,
@@ -610,10 +606,16 @@ fn achat_completions(
 #[pyfunction]
 fn gil_stats(py: Python<'_>) -> PyResult<Py<PyAny>> {
     let stats = PyDict::new(py);
-    stats.set_item("releases", gil::release_count())?;
+    stats.set_item("releases", release_count())?;
     Ok(stats.into_any().unbind())
 }
 
+#[cfg(feature = "panic-test")]
+#[pyfunction]
+fn _panic_for_test() {
+    panic!("intentional PyO3 panic smoke test");
+}
+
 #[pymodule]
 fn _native(module: &Bound<'_, PyModule>) -> PyResult<()> {
     let py = module.py();
@@ -630,5 +632,7 @@ fn _native(module: &Bound<'_, PyModule>) -> PyResult<()> {
     module.add_function(wrap_pyfunction!(achat_completions, module)?)?;
     module.add_class::<ResponsesWebSocketConnection>()?;
     module.add_function(wrap_pyfunction!(gil_stats, module)?)?;
+    #[cfg(feature = "panic-test")]
+    module.add_function(wrap_pyfunction!(_panic_for_test, module)?)?;
     Ok(())
 }

litellm-rust/crates/python-bridge/tests/marshal_boundary.rs

@@ -1,7 +1,7 @@
 use std::fs;
 use std::path::{Path, PathBuf};
 
-const DISALLOWED_OUTSIDE_MARSHAL: &[&str] = &[
+const DISALLOWED_OUTSIDE_INTEROP: &[&str] = &[
     "py.import(\"json\")",
     "pythonize::",
     "serde_json::to_string",
@@ -33,18 +33,15 @@ fn rust_sources(directory: &Path) -> Vec<PathBuf> {
 }
 
 #[test]
-fn serialization_is_centralized_in_marshal_module() {
+fn serialization_uses_the_interop_boundary() {
     let root = source_root();
 
     for path in rust_sources(&root) {
-        if path == root.join("marshal.rs") {
-            continue;
-        }
         let source = fs::read_to_string(&path).expect("bridge source should be readable");
-        for disallowed in DISALLOWED_OUTSIDE_MARSHAL {
+        for disallowed in DISALLOWED_OUTSIDE_INTEROP {
             assert!(
                 !source.contains(disallowed),
-                "{} bypasses the typed marshal module with `{disallowed}`",
+                "{} bypasses litellm-python-interop with `{disallowed}`",
                 path.display()
             );
         }

litellm-rust/crates/python-interop/AGENTS.md

@@ -0,0 +1 @@
+litellm-python-interop is the domain-neutral PyO3 foundation. Keep generic Python/Serde conversion and interpreter primitives here. Do not add LiteLLM domain crates, route types, API registration, or cdylib build features.

litellm-rust/crates/python-interop/Cargo.toml

@@ -0,0 +1,15 @@
+[package]
+name = "litellm-python-interop"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+repository.workspace = true
+
+[dependencies]
+pyo3.workspace = true
+pythonize.workspace = true
+serde.workspace = true
+
+[dev-dependencies]
+rstest.workspace = true
+serde_json.workspace = true

litellm-rust/crates/python-interop/src/gil.rs

@@ -0,0 +1,21 @@
+use std::sync::atomic::{AtomicU64, Ordering};
+
+use pyo3::prelude::*;
+
+static GIL_RELEASES: AtomicU64 = AtomicU64::new(0);
+
+/// Runs work detached from the interpreter and records the release.
+///
+/// `f` must not access Python state while the interpreter is detached.
+pub fn release_gil<T, F>(py: Python<'_>, f: F) -> T
+where
+    F: FnOnce() -> T + Send,
+    T: Send,
+{
+    GIL_RELEASES.fetch_add(1, Ordering::Relaxed);
+    py.detach(f)
+}
+
+pub fn release_count() -> u64 {
+    GIL_RELEASES.load(Ordering::Relaxed)
+}

litellm-rust/crates/python-interop/src/lib.rs

@@ -0,0 +1,5 @@
+mod gil;
+mod marshal;
+
+pub use gil::{release_count, release_gil};
+pub use marshal::{from_py, to_py};

litellm-rust/crates/python-interop/tests/interop.rs

@@ -0,0 +1,44 @@
+use pyo3::Python;
+use rstest::{fixture, rstest};
+use serde_json::{Value, json};
+
+use litellm_python_interop::{from_py, release_count, release_gil, to_py};
+
+struct InitializedPython;
+
+impl InitializedPython {
+    fn attach<F, R>(&self, f: F) -> R
+    where
+        F: for<'py> FnOnce(Python<'py>) -> R,
+    {
+        Python::attach(f)
+    }
+}
+
+#[fixture]
+#[once]
+fn initialized_python() -> InitializedPython {
+    Python::initialize();
+    InitializedPython
+}
+
+#[rstest]
+fn serde_values_round_trip_through_python(#[from(initialized_python)] python: &InitializedPython) {
+    python.attach(|py| {
+        let expected = json!({"model": "test", "items": [1, true, null]});
+        let python_value = to_py(py, &expected).expect("value should convert to Python");
+        let actual: Value =
+            from_py(python_value.bind(py)).expect("Python value should convert to serde");
+
+        assert_eq!(actual, expected);
+    });
+}
+
+#[rstest]
+fn release_gil_runs_work_and_records_it(#[from(initialized_python)] python: &InitializedPython) {
+    let before = release_count();
+    let result = python.attach(|py| release_gil(py, || 42));
+
+    assert_eq!(result, 42);
+    assert_eq!(release_count(), before + 1);
+}

litellm/__init__.py

@@ -29,7 +29,7 @@ def _dev_env_hot_reload_enabled() -> bool:
 if os.getenv("LITELLM_MODE", "DEV") == "DEV":
     _dotenv.load_dotenv(override=_dev_env_hot_reload_enabled())
 
-from collections.abc import Sequence
+from collections.abc import Mapping, Sequence
 from typing import (
     Any,
     Callable,
@@ -490,6 +490,7 @@ public_mcp_hub_strict_whitelist: bool = True
 public_model_groups: Optional[List[str]] = None
 public_agent_groups: Optional[List[str]] = None
 agent_search_embedding_model: Optional[str] = None
+mcp_tool_search: Optional[Mapping[str, object]] = None
 # Supports both old format (Dict[str, str]) and new format (Dict[str, Dict[str, Any]])
 # New format: { "displayName": { "url": "...", "index": 0 } }
 # Old format: { "displayName": "url" } (for backward compatibility)

litellm/_logging.py

@@ -6,6 +6,7 @@ import sys
 from datetime import datetime
 from logging import Formatter
 from typing import Any, Final, TextIO
+from urllib.parse import unquote
 
 import litellm
 from litellm.constants import (
@@ -146,6 +147,72 @@ class SecretRedactionFilter(logging.Filter):
 _secret_filter: Final = SecretRedactionFilter()
 
 
+_MAX_SCRUBBED_ACCESS_ARG: Final = 512
+
+_REDACTION_PLACEHOLDER: Final = "REDACTED"
+
+
+def _hides_a_credential(value: str) -> bool:
+    """Whether *value* only looks clean until it is percent-decoded."""
+    decoded: Final = unquote(value)
+    return _redact_string(decoded) != decoded
+
+
+def _drop_encoded_credential(scrubbed: str) -> str:
+    """Drop the part of a request target that only decoding shows to be a secret.
+
+    The request parser decodes query names and values, so `?k%65y=sk%2D...` is a
+    working credential that the patterns, which match literal text, do not see.
+    The decoded text is never logged back: it can carry a newline, and forging
+    log lines is not a trade worth making for a readable request target.
+    """
+    path, separator, _query = scrubbed.partition("?")
+    if _hides_a_credential(path):
+        return _REDACTION_PLACEHOLDER
+    if separator and _hides_a_credential(scrubbed):
+        return f"{path}?{_REDACTION_PLACEHOLDER}"
+    return scrubbed
+
+
+def _scrub_access_arg(value: str) -> str:
+    """Redact one access-log positional arg, bounding the scanned length.
+
+    The request target is the only input to the secret regex an unauthenticated
+    caller controls end to end, so it is cut back to a whole query parameter
+    before it is scanned; a half-parameter would be too short to match its
+    pattern and would then be logged raw.
+    """
+    if len(value) <= _MAX_SCRUBBED_ACCESS_ARG:
+        return _drop_encoded_credential(_redact_string(value))
+    head: Final = value[:_MAX_SCRUBBED_ACCESS_ARG]
+    kept: Final = head[: max(head.rfind("?"), head.rfind("&"))] if "?" in head else head
+    scrubbed: Final = _drop_encoded_credential(_redact_string(kept))
+    return f"{scrubbed}... ({len(value) - len(kept)} more chars truncated) ..."
+
+
+class AccessLogRedactionFilter(logging.Filter):
+    """Scrubs known secret/credential patterns from HTTP access-log records.
+
+    uvicorn's AccessFormatter unpacks ``record.args`` as a five-element tuple at
+    emit time, so SecretRedactionFilter cannot be reused here: it collapses the
+    record into ``record.msg`` and clears the args, and the formatter then raises.
+    """
+
+    def filter(self, record: logging.LogRecord) -> bool:
+        if not _ENABLE_SECRET_REDACTION:
+            return True
+        if isinstance(record.args, tuple) and record.args:
+            record.args = tuple(  # rebind-ok: a Filter scrubs records in place
+                _scrub_access_arg(arg) if isinstance(arg, str) else arg for arg in record.args
+            )
+            return True
+        # No positional args means everything is in msg, where collapsing is correct.
+        return _secret_filter.filter(record)
+
+
+_access_log_filter: Final = AccessLogRedactionFilter()
+
+
 def _get_max_string_length_stdout_log() -> int:
     """Read the limit per record so a value loaded later via proxy config
     environment_variables is honored."""
@@ -553,6 +620,14 @@ _REDACTED_THIRD_PARTY_LOGGERS: Final[tuple[str, ...]] = (
     "uvicorn.error",
 )
 
+# Access loggers, which emit the full request target, so a credential passed as a
+# query parameter (e.g. `/key/info?key=`) lands on stdout verbatim. uvicorn.access
+# covers uvicorn.run, --run_gunicorn (its worker_class is UvicornWorker, so the
+# access line is still uvicorn's) and an embedding host app. --run_hypercorn and
+# --run_granian log through their own loggers in their own record shapes, and
+# both ship with access logging off.
+_REDACTED_ACCESS_LOGGERS: Final[tuple[str, ...]] = ("uvicorn.access",)
+
 
 def _redact_third_party_loggers() -> None:
     """Extend secret redaction to records litellm does not emit directly.
@@ -575,6 +650,8 @@ def _redact_third_party_loggers() -> None:
     """
     for name in _REDACTED_THIRD_PARTY_LOGGERS:
         logging.getLogger(name).addFilter(_secret_filter)
+    for name in _REDACTED_ACCESS_LOGGERS:
+        logging.getLogger(name).addFilter(_access_log_filter)
 
 
 # Call the suppression function

litellm/constants.py

@@ -1742,6 +1742,7 @@ LITELLM_SETTINGS_SAFE_DB_OVERRIDES: Final = [
     "anthropic_prompt_caching_ttl",
     "max_ui_session_budget",
     "budget_rollover",
+    "mcp_tool_search",
 ]
 SPECIAL_LITELLM_AUTH_TOKEN: Final = ["ui-token"]
 DEFAULT_MANAGEMENT_OBJECT_IN_MEMORY_CACHE_TTL = int(os.getenv("DEFAULT_MANAGEMENT_OBJECT_IN_MEMORY_CACHE_TTL", 60))

litellm/containers/main.py

@@ -1,7 +1,7 @@
 import asyncio
 import contextvars
 import json
-from collections.abc import Coroutine, Mapping
+from collections.abc import Callable, Coroutine, Mapping
 from functools import partial
 from typing import Final, Literal, overload
 
@@ -47,6 +47,13 @@ __all__ = [
 
 
 ##### Container Create #######################
+async def _encode_created_container_id(
+    pending: Coroutine[object, object, ContainerObject],
+    encode: Callable[[ContainerObject], ContainerObject],
+) -> ContainerObject:
+    return encode(await pending)
+
+
 @client
 async def acreate_container(
     name: str,
@@ -256,16 +263,16 @@ def create_container(
             _is_async=_is_async,
         )
 
-        # Encode container_id with provider/model metadata for routing
+        encode: Final = partial(
+            ContainerRequestUtils.encode_container_id_in_response,
+            custom_llm_provider=custom_llm_provider,
+            litellm_metadata=kwargs.get("litellm_metadata"),
+            extra_body=extra_body,
+        )
         if isinstance(container_obj, ContainerObject):
-            container_obj = ContainerRequestUtils.encode_container_id_in_response(
-                response_obj=container_obj,
-                custom_llm_provider=custom_llm_provider,
-                litellm_metadata=kwargs.get("litellm_metadata"),
-                extra_body=extra_body,
-            )
+            return encode(container_obj)
 
-        return container_obj
+        return _encode_created_container_id(pending=container_obj, encode=encode)
 
     except Exception as e:
         raise litellm.exception_type(

litellm/integrations/anthropic_cache_control_hook.py

@@ -755,6 +755,7 @@ class AnthropicCacheControlHook(CustomPromptManagement):
     def record_gateway_injection(
         request_kwargs: Mapping[str, object],
         added: int,
+        injected_for_every_deployment: bool = False,
     ) -> None:
         """Name the deployment whose payload the gateway, not the client, put breakpoints on.
 
@@ -771,7 +772,16 @@ class AnthropicCacheControlHook(CustomPromptManagement):
 
         A pass that runs before a deployment is chosen, which is what the proxy does for
         prompt templates, injects into the payload every leg goes on to send, so it marks
-        the request for all of them rather than for one.
+        the request for all of them rather than for one. Such a pass says so with
+        ``injected_for_every_deployment`` instead of relying on the shape of
+        ``request_kwargs``: the router's prompt-management factory stamps a provisional
+        deployment's ``model_info`` into kwargs before the prompt pass runs, and billing
+        the request through any other deployment would silently drop the credit. An
+        every-deployment mark, once written, also never narrows: a later per-leg stamp
+        (the Bedrock converse tool_config one included) describes one leg of a payload
+        every leg sends, so narrowing to it would uncredit whichever leg gets billed
+        after a failover. Both losses are fail-closed under-crediting, which is why the
+        guard only protects the sentinel and per-leg marks still overwrite each other.
 
         Only what this pass actually placed counts. A ``tool_config`` point is placed by
         the Bedrock converse transform, and only when the request carries tools, so the
@@ -801,13 +811,19 @@ class AnthropicCacheControlHook(CustomPromptManagement):
             ),
             None,
         )
-        if bucket is not None:
-            model_info: Final = request_kwargs.get("model_info")
-            bucket[GATEWAY_INJECTED_CACHE_METADATA_KEY] = (
-                model_info.get("id", GATEWAY_INJECTED_FOR_EVERY_DEPLOYMENT)
-                if isinstance(model_info, dict)
-                else GATEWAY_INJECTED_FOR_EVERY_DEPLOYMENT
-            )
+        if bucket is None:
+            return
+        if bucket.get(GATEWAY_INJECTED_CACHE_METADATA_KEY) == GATEWAY_INJECTED_FOR_EVERY_DEPLOYMENT:
+            return
+        if injected_for_every_deployment:
+            bucket[GATEWAY_INJECTED_CACHE_METADATA_KEY] = GATEWAY_INJECTED_FOR_EVERY_DEPLOYMENT
+            return
+        model_info: Final = request_kwargs.get("model_info")
+        bucket[GATEWAY_INJECTED_CACHE_METADATA_KEY] = (
+            model_info.get("id", GATEWAY_INJECTED_FOR_EVERY_DEPLOYMENT)
+            if isinstance(model_info, dict)
+            else GATEWAY_INJECTED_FOR_EVERY_DEPLOYMENT
+        )
 
     @staticmethod
     def maybe_inject_cache_control(

litellm/integrations/otel/langfuse_logger.py

@@ -0,0 +1,60 @@
+from collections.abc import AsyncGenerator, AsyncIterator, Callable, Mapping
+from typing import TYPE_CHECKING, Final
+
+from litellm._logging import verbose_logger
+from litellm.integrations.otel.logger import OpenTelemetryV2
+from litellm.integrations.otel.mappers.langfuse import LANGFUSE_OBSERVATION_INPUT, LANGFUSE_OBSERVATION_OUTPUT
+from litellm.integrations.otel.model.request_io import request_input, response_output, stream_output
+from litellm.integrations.otel.plumbing.context import request_root_span
+
+if TYPE_CHECKING:
+    from litellm.proxy._types import UserAPIKeyAuth
+    from litellm.types.utils import ModelResponseStream
+
+
+class LangfuseOpenTelemetryV2(OpenTelemetryV2):
+    """Stamps the request's input and output on the root observation while it is still recording.
+
+    Langfuse shows a trace's input and output from its root observation. The proxy's root span ends
+    when the response is sent, before the success callback runs, so both stamps come from the
+    post-call hooks in the request task: the request as it stands after the pre-call chain and the
+    response as it is returned, for the call types whose response renders as a message.
+    """
+
+    async def async_post_call_success_hook(
+        self,
+        data: Mapping[str, object],
+        user_api_key_dict: "UserAPIKeyAuth",
+        response: object,
+    ) -> None:
+        self._stamp_root_io(data, lambda: response_output(response))
+
+    async def async_post_call_streaming_iterator_hook(
+        self,
+        user_api_key_dict: "UserAPIKeyAuth",
+        response: "AsyncIterator[ModelResponseStream]",
+        request_data: Mapping[str, object],
+    ) -> "AsyncGenerator[ModelResponseStream, None]":
+        relayed: Final[list[ModelResponseStream]] = []  # mutable-ok: relayed as they arrive, assembled at end of stream
+        async for chunk in response:
+            relayed.append(chunk)
+            yield chunk
+        self._stamp_root_io(request_data, lambda: stream_output(tuple(relayed), request_data))
+
+    def _stamp_root_io(self, data: Mapping[str, object], render_output: Callable[[], str | None]) -> None:
+        root: Final = request_root_span()
+        if root is None or not root.is_recording():
+            return
+        try:
+            output: Final = render_output()
+            if output is None:
+                return
+            root.set_attribute(LANGFUSE_OBSERVATION_OUTPUT, output)
+            rendered_input: Final = request_input(data)
+        except Exception:  # noqa: BLE001  # telemetry must never fail the request it describes
+            verbose_logger.debug(
+                "otel v2 langfuse: could not render the root observation input or output", exc_info=True
+            )
+            return
+        if rendered_input is not None:
+            root.set_attribute(LANGFUSE_OBSERVATION_INPUT, rendered_input)

litellm/integrations/otel/logger.py

@@ -4,6 +4,7 @@ from collections import OrderedDict
 from collections.abc import Callable, Iterator, Mapping, Sequence
 from contextlib import contextmanager
 from datetime import datetime
+from types import MappingProxyType
 from typing import TYPE_CHECKING, Any, Final, cast
 
 from opentelemetry.context import Context, attach, get_current
@@ -909,3 +910,29 @@ def phase_span(name: str) -> "Iterator[Span | None]":
         return
     with logger.start_phase_span(name) as span:
         yield span
+
+
+def build_otel_v2_logger(
+    config: OpenTelemetryV2Config,
+    callback_name: str | None = None,
+    tracer_provider: TracerProvider | None = None,
+    logger_provider: LoggerProvider | None = None,
+    meter_provider: "MeterProvider | None" = None,
+    settings: Mapping[str, object] = MappingProxyType({}),
+) -> OpenTelemetryV2:
+    return _logger_class(config)(
+        config=config,
+        callback_name=callback_name,
+        tracer_provider=tracer_provider,
+        logger_provider=logger_provider,
+        meter_provider=meter_provider,
+        **settings,
+    )
+
+
+def _logger_class(config: OpenTelemetryV2Config) -> type[OpenTelemetryV2]:
+    if "langfuse" not in config.mapper_names or not config.capture_span_content:
+        return OpenTelemetryV2
+    from litellm.integrations.otel.langfuse_logger import LangfuseOpenTelemetryV2
+
+    return LangfuseOpenTelemetryV2

litellm/integrations/otel/mappers/langfuse.py

@@ -11,6 +11,7 @@ the JSON-serialized payloads. ``_llm_call`` just applies both tables.
 
 import json
 from collections.abc import Callable
+from typing import Final
 
 from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue, SpanData
 from litellm.integrations.otel.mappers.utils import (
@@ -25,6 +26,9 @@ from litellm.integrations.otel.model.payloads import (
     LLMUsage,
 )
 
+LANGFUSE_OBSERVATION_INPUT: Final = "langfuse.observation.input"
+LANGFUSE_OBSERVATION_OUTPUT: Final = "langfuse.observation.output"
+
 
 class LangfuseMapper:
     _LLM_CALL_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = {
@@ -56,8 +60,8 @@ class LangfuseMapper:
         "langfuse.observation.model.parameters": lambda d: json_if(
             collect(LangfuseMapper._MODEL_PARAMS, d.request_params)
         ),
-        "langfuse.observation.input": lambda d: serialize_messages(d.messages_in),
-        "langfuse.observation.output": lambda d: serialize_messages(output_messages(d)),
+        LANGFUSE_OBSERVATION_INPUT: lambda d: serialize_messages(d.messages_in),
+        LANGFUSE_OBSERVATION_OUTPUT: lambda d: serialize_messages(output_messages(d)),
         "langfuse.observation.usage_details": lambda d: json_if(collect(LangfuseMapper._USAGE_FIELDS, d.usage)),
         "langfuse.observation.cost_details": lambda d: (
             json.dumps({"total": d.response_cost}) if d.response_cost is not None else None

litellm/integrations/otel/model/request_io.py

@@ -0,0 +1,90 @@
+from collections.abc import Mapping, Sequence
+from typing import Final, Literal
+
+from pydantic import BaseModel, ConfigDict, Field, TypeAdapter, ValidationError
+from typing_extensions import ReadOnly, TypedDict
+
+import litellm
+from litellm.integrations.otel.mappers.utils import json_or_none
+from litellm.proxy.guardrails.anthropic_sse import assemble_anthropic_sse_stream, is_raw_sse_stream
+from litellm.types.llms.openai import ResponseCompletedEvent, ResponsesAPIResponse
+from litellm.types.utils import ModelResponse, ModelResponseStream
+
+_SYSTEM_KEYS: Final = ("system", "instructions")
+_TURNS: Final = TypeAdapter(tuple[object, ...])
+_MESSAGES: Final = TypeAdapter(list[object] | None)
+
+
+class _Turn(TypedDict):
+    role: ReadOnly[str]
+    content: ReadOnly[object]
+
+
+class _AnthropicMessage(BaseModel):
+    model_config = ConfigDict(frozen=True)
+
+    type: Literal["message"] = Field(exclude=True)
+    role: str = "assistant"
+    content: object = None
+
+
+def request_input(data: Mapping[str, object]) -> str | None:
+    turns: Final = data.get("messages", data.get("input"))
+    if turns is None:
+        return None
+    return json_or_none((*_system_turns(data), *_user_turns(turns)))
+
+
+def _system_turns(data: Mapping[str, object]) -> tuple[_Turn, ...]:
+    return tuple(_Turn(role="system", content=data[key]) for key in _SYSTEM_KEYS if data.get(key) is not None)
+
+
+def _user_turns(turns: object) -> tuple[object, ...]:
+    if isinstance(turns, str):
+        return (_Turn(role="user", content=turns),)
+    try:
+        return _TURNS.validate_python(turns)
+    except ValidationError:
+        return (_Turn(role="user", content=turns),)
+
+
+def response_output(response: object) -> str | None:
+    match response:
+        case ModelResponse():
+            return json_or_none(tuple(choice.message.model_dump(exclude_none=True) for choice in response.choices))
+        case ResponsesAPIResponse():
+            return json_or_none(response.model_dump(exclude_none=True).get("output"))
+        case _:
+            return _anthropic_message_output(response)
+
+
+def _anthropic_message_output(message: object) -> str | None:
+    try:
+        parsed: Final = _AnthropicMessage.model_validate(message)
+    except ValidationError:
+        return None
+    return json_or_none((parsed.model_dump(),))
+
+
+def stream_output(chunks: Sequence[object], data: Mapping[str, object]) -> str | None:
+    if not chunks:
+        return None
+    if is_raw_sse_stream(chunks):
+        return response_output(assemble_anthropic_sse_stream(chunks))
+    if all(isinstance(chunk, ModelResponseStream) for chunk in chunks):
+        return response_output(_assembled_chat_stream(chunks, data))
+    return response_output(_completed_response(chunks))
+
+
+def _assembled_chat_stream(chunks: Sequence[object], data: Mapping[str, object]) -> object:
+    try:
+        return litellm.stream_chunk_builder(  # pyright: ignore[reportUnknownMemberType]  # upstream types chunks as a bare list
+            chunks=list(chunks),  # mutable-ok: stream_chunk_builder takes a list
+            messages=_MESSAGES.validate_python(data.get("messages")),
+        )
+    except (litellm.APIError, ValidationError):
+        return None
+
+
+def _completed_response(chunks: Sequence[object]) -> ResponsesAPIResponse | None:
+    return next((chunk.response for chunk in reversed(chunks) if isinstance(chunk, ResponseCompletedEvent)), None)

litellm/litellm_core_utils/litellm_logging.py

@@ -901,6 +901,7 @@ class Logging(LiteLLMLoggingBaseClass):
         prompt_label: str | None = None,
         prompt_version: int | None = None,
         request_kwargs: dict[str, object] | None = None,  # mutable-ok: marker stamped into live request kwargs
+        injected_for_every_deployment: bool = False,
     ) -> tuple[str, list[AllMessageValues], dict]:
         from litellm.integrations.anthropic_cache_control_hook import AnthropicCacheControlHook
 
@@ -933,6 +934,7 @@ class Logging(LiteLLMLoggingBaseClass):
                 AnthropicCacheControlHook.record_gateway_injection(
                     request_kwargs,
                     AnthropicCacheControlHook.count_request_cache_breakpoints(messages) - breakpoints_before,
+                    injected_for_every_deployment=injected_for_every_deployment,
                 )
         self.messages = messages
         return model, messages, non_default_params
@@ -950,6 +952,7 @@ class Logging(LiteLLMLoggingBaseClass):
         prompt_label: str | None = None,
         prompt_version: int | None = None,
         request_kwargs: dict[str, object] | None = None,  # mutable-ok: marker stamped into live request kwargs
+        injected_for_every_deployment: bool = False,
     ) -> tuple[str, list[AllMessageValues], dict]:
         from litellm.integrations.anthropic_cache_control_hook import AnthropicCacheControlHook
 
@@ -985,6 +988,7 @@ class Logging(LiteLLMLoggingBaseClass):
                 AnthropicCacheControlHook.record_gateway_injection(
                     request_kwargs,
                     AnthropicCacheControlHook.count_request_cache_breakpoints(messages) - breakpoints_before,
+                    injected_for_every_deployment=injected_for_every_deployment,
                 )
         self.messages = messages
         return model, messages, non_default_params
@@ -4390,13 +4394,15 @@ def _init_custom_logger_compatible_class(
             from litellm.integrations.otel.model.config import is_otel_v2_enabled
 
             if is_otel_v2_enabled():
-                from litellm.integrations.otel.logger import OpenTelemetryV2
+                from litellm.integrations.otel.logger import OpenTelemetryV2, build_otel_v2_logger
+                from litellm.integrations.otel.model.config import OpenTelemetryV2Config
 
                 for callback in _in_memory_loggers:
-                    if type(callback) is OpenTelemetryV2:
+                    if isinstance(callback, OpenTelemetryV2):
                         return callback
-                otel_logger_v2: Final = OpenTelemetryV2(
-                    **_get_custom_logger_settings_from_proxy_server(callback_name=logging_integration)
+                otel_settings: Final = _get_custom_logger_settings_from_proxy_server(callback_name=logging_integration)
+                otel_logger_v2: Final = build_otel_v2_logger(
+                    config=OpenTelemetryV2Config(**otel_settings), settings=otel_settings
                 )
                 _in_memory_loggers.append(otel_logger_v2)
                 _maybe_auto_initialize_arize_phoenix(_in_memory_loggers)
@@ -4759,7 +4765,7 @@ def _maybe_construct_otel_v2(callback_name: str, _in_memory_loggers: list[Custom
 
     if not is_otel_v2_enabled():
         return None
-    from litellm.integrations.otel.logger import OpenTelemetryV2
+    from litellm.integrations.otel.logger import OpenTelemetryV2, build_otel_v2_logger
     from litellm.integrations.otel.presets import PRESET_BY_CALLBACK
 
     preset_fn: Final = PRESET_BY_CALLBACK.get(callback_name)
@@ -4774,7 +4780,7 @@ def _maybe_construct_otel_v2(callback_name: str, _in_memory_loggers: list[Custom
         # If env vars are missing or the preset raises, defer to the legacy path
         # so customers get the same error story they had before V2 landed.
         return None
-    v2_logger: Final = OpenTelemetryV2(config=config, callback_name=callback_name)
+    v2_logger: Final = build_otel_v2_logger(config=config, callback_name=callback_name)
     _in_memory_loggers.append(v2_logger)
     return v2_logger
 

litellm/litellm_core_utils/secret_redaction.py

@@ -30,6 +30,11 @@ def _build_secret_patterns() -> "re.Pattern[str]":
         r"Basic\s+[A-Za-z0-9+/]{10,}={0,2}",
         # OpenAI / Anthropic sk- prefixed keys
         rf"sk-[A-Za-z0-9\-_]{{{MINIMUM_CUSTOM_KEY_LENGTH - len('sk-')},}}",
+        # Credentials passed as URL query params. Terminated by "&" like the key=
+        # and sig= patterns below, so the rest of the request line survives in an
+        # access log. Must precede the generic patterns to win at the same position.
+        r"(?<=[?&])(?:api[_-]?key|\w*(?:token|password|passwd|client_secret|secret_key|_secret))"
+        r"=[^\s&'\"]+",
         # Generic api_key / api-key / apikey (handles 'key': 'value' dict repr)
         r"(?:api[_-]?key)['\"]?\s*[:=]\s*['\"]?[^\s,'\"})\]{}>]{8,}",
         # x-api-key / api-key header values (handles 'key': 'value' dict repr)
@@ -45,8 +50,10 @@ def _build_secret_patterns() -> "re.Pattern[str]":
         # Word boundary prevents O(n^2) backtracking on long word-char runs.
         r"(?:^|(?<=\W))\w*(?:password|passwd|client_secret|secret_key|_secret)"
         r"['\"]?\s*[:=]\s*['\"]?[^\s,'\"})\]{}>]+",
-        # Database connection string credentials (scheme://user:pass@host)
-        r"(?<=://)[^\s'\"]*:[^\s'\"@]+(?=@)",
+        # Database connection string credentials (scheme://user:pass@host).
+        # The user half stops at the ":" separator and both halves are length-capped,
+        # so a long attacker-supplied URL cannot backtrack quadratically.
+        r"(?<=://)[^\s'\":]{0,4096}:[^\s'\"]{1,4096}(?=@)",
         # Databricks personal access tokens
         r"dapi[0-9a-f]{32}",
         # Module-level provider keys logged as litellm.<provider>_key=<value>
@@ -67,8 +74,10 @@ def _build_secret_patterns() -> "re.Pattern[str]":
         r"""['\"]?\s*[:=]\s*['\"]?[^\s,'\"})\]{}>]+""",
         # Raw JWTs (without Bearer prefix)
         r"\beyJ[A-Za-z0-9_-]{10,}\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]*",
-        # Azure SAS tokens in URLs
-        r"[?&]sig=[A-Za-z0-9%+/=]+",
+        # Azure SAS tokens in URLs. The delimiter is a lookbehind, like the
+        # `key=` pattern above, so the `?` or `&` survives and the redacted URL
+        # stays well formed (this string is often a request line in a log).
+        r"(?<=[?&])sig=[A-Za-z0-9%+/=]+",
         # Full JSON service-account blobs (single-line and multi-line)
         r'\{[^{}]*"type"\s*:\s*"service_account"[^{}]*(?:\{[^{}]*\}[^{}]*)*\}',
     ]

litellm/llms/base_llm/files/azure_blob_storage_backend.py

@@ -7,16 +7,36 @@ to reuse all authentication and Azure Storage operations.
 """
 
 import time
+from pathlib import Path
 from typing import Final
 from urllib.parse import quote, urlparse
 
 from litellm._logging import verbose_logger
 from litellm._uuid import uuid
 from litellm.integrations.azure_storage.azure_storage import AzureBlobStorageLogger
+from litellm.proxy.common_utils.path_utils import safe_filename
 
 from .storage_backend import BaseFileStorageBackend
 
 
+def _safe_basename(original_filename: str) -> str:
+    try:
+        return safe_filename(original_filename)
+    except ValueError:
+        return "file"
+
+
+def _safe_extension(original_filename: str) -> str:
+    """The extension off a basename, with no path separators or traversal sequences.
+
+    original_filename.split(".")[-1] does not parse path structure, so a filename
+    like "a.jsonl/../../etc/cron.d/x" would put "../../etc/cron.d/x" straight into
+    the blob path built below. Path.suffix only ever looks at the last path
+    component, so routing through safe_filename() first closes that off.
+    """
+    return Path(_safe_basename(original_filename)).suffix.lstrip(".")
+
+
 class AzureBlobStorageBackend(BaseFileStorageBackend, AzureBlobStorageLogger):
     """
     Azure Blob Storage backend implementation.
@@ -81,16 +101,15 @@ class AzureBlobStorageBackend(BaseFileStorageBackend, AzureBlobStorageLogger):
     def _generate_file_name(self, original_filename: str, file_naming_strategy: str) -> str:
         """Generate file name based on naming strategy."""
         if file_naming_strategy == "original_filename":
-            # Use original filename, but sanitize it
-            return quote(original_filename, safe="")
+            return quote(_safe_basename(original_filename), safe="")
         elif file_naming_strategy == "timestamp":
             # Use timestamp
-            extension = original_filename.split(".")[-1] if "." in original_filename else ""
+            extension = _safe_extension(original_filename)
             timestamp: Final = int(time.time() * 1000)  # milliseconds
             return f"{timestamp}.{extension}" if extension else str(timestamp)
         else:  # default to "uuid"
             # Use UUID
-            extension = original_filename.split(".")[-1] if "." in original_filename else ""
+            extension = _safe_extension(original_filename)
             file_uuid: Final = str(uuid.uuid4())
             return f"{file_uuid}.{extension}" if extension else file_uuid
 

litellm/main.py

@@ -141,7 +141,6 @@ from litellm.utils import (
     convert_to_model_response_object,
     create_pretrained_tokenizer,
     create_tokenizer,
-    get_api_key,
     get_llm_provider,
     get_model_info,
     get_non_default_completion_params,

litellm/proxy/_experimental/mcp_server/auth/user_api_key_auth_mcp.py

@@ -1149,10 +1149,11 @@ class MCPRequestHandler:
         would miss a real outage wrapped inside it."""
         from litellm.proxy.db.exception_handler import PrismaDBExceptionHandler
 
-        if PrismaDBExceptionHandler.is_database_service_unavailable_error_in_chain(e):
+        outage: Final = PrismaDBExceptionHandler.find_database_service_unavailable_error_in_chain(e)
+        if outage is not None:
             raise HTTPException(
                 status_code=503,
-                detail="Service Unavailable, the authentication database is temporarily unreachable. Please retry shortly.",
+                detail=PrismaDBExceptionHandler.database_unavailable_message(outage),
             ) from None
 
     @staticmethod

litellm/proxy/_experimental/mcp_server/bridge_token_flow.py

@@ -101,18 +101,29 @@ class _ResolvedKey:
     key: "UserAPIKeyAuth"
 
 
-_KeyResolutionFailure = Literal["no_active_key", "unavailable", "unresolvable"]
+_KeyResolutionFailure = Literal["no_active_key", "unavailable", "faulted", "unresolvable"]
 """Why a token request yielded no active litellm key, kept distinct so a caller statuses each truthfully
 instead of blaming the client for a gateway problem:
 - ``no_active_key``: none was presented, or the presented key is unknown / blocked / expired (the
   caller's request is at fault)
 - ``unavailable``: the auth database was transiently unreachable while resolving (retryable)
+- ``faulted``: the auth database's query engine reported a fault that retrying will not clear (still a
+  503, but the wording must not tell the operator to wait)
 - ``unresolvable``: the gateway cannot resolve identity right now (no DB connection, or an unexpected
   error) -- a gateway fault, not the caller's
 The classification mirrors admission's ``_reload_admitted_key`` so the mint (ingress) and admission
 (egress) never disagree on the status of the same outage."""
 
 
+def _database_failure(exc: Exception) -> Literal["unavailable", "faulted"]:
+    from litellm.proxy.db.exception_handler import (  # noqa: PLC0415  # inline import avoids a module-load circular import
+        PrismaDBExceptionHandler,
+    )
+
+    fault: Final = PrismaDBExceptionHandler.find_database_service_unavailable_error_in_chain(exc) or exc
+    return "faulted" if PrismaDBExceptionHandler.is_permanent_database_fault(fault) else "unavailable"
+
+
 async def _resolve_active_litellm_key(request: Request) -> "_ResolvedKey | _KeyResolutionFailure":
     """Resolve the presented litellm key to an active key record, or say precisely why not.
 
@@ -170,7 +181,7 @@ async def _reload_active_key_by_hash(key_hash: str) -> "_ResolvedKey | _KeyResol
         return "no_active_key"
     except Exception as exc:  # noqa: BLE001  # classify: a DB outage is retryable, anything else is an opaque gateway fault
         if PrismaDBExceptionHandler.is_database_service_unavailable_error(exc):
-            return "unavailable"
+            return _database_failure(exc)
         verbose_logger.debug(
             "_reload_active_key_by_hash: unexpected key-resolution error (%s)",
             type(exc).__name__,
@@ -225,8 +236,9 @@ async def load_active_user_by_id(user_id: str) -> "LiteLLM_UserTable | _KeyResol
     except (ProxyException, HTTPException):
         return "no_active_key"
     except Exception as exc:  # noqa: BLE001  # a DB outage is retryable; a missing user (get_user_object's wrapped ValueError) or any other resolution failure fails closed as no_active_key, never a 500
-        if PrismaDBExceptionHandler.is_database_service_unavailable_error_in_chain(exc):
-            return "unavailable"
+        outage: Final = PrismaDBExceptionHandler.find_database_service_unavailable_error_in_chain(exc)
+        if outage is not None:
+            return _database_failure(outage)
         verbose_logger.debug("_reload_active_user_by_id: user-resolution error (%s)", type(exc).__name__)
         return "no_active_key"
     if user_object is None:
@@ -383,6 +395,7 @@ _BridgeMintError = Literal[
     "no_identity",
     "invalid_refresh",
     "identity_unavailable",
+    "identity_faulted",
     "identity_unresolvable",
     "not_configured",
     "no_upstream_token",
@@ -433,6 +446,13 @@ def _bridge_mint_error_response(error: _BridgeMintError) -> JSONResponse:
                 "temporarily_unavailable",
                 "the authentication database is temporarily unreachable; retry shortly",
             )
+        case "identity_faulted":
+            status, code, desc = (
+                503,
+                "temporarily_unavailable",
+                "the authentication database reported a fault that is not a transient outage; "
+                "retrying will not help until the gateway deployment is repaired",
+            )
         case "identity_unresolvable":
             status, code, desc = (
                 500,
@@ -485,6 +505,8 @@ def _key_resolution_failure_to_mint_error(failure: _KeyResolutionFailure) -> _Br
             return "no_identity"
         case "unavailable":
             return "identity_unavailable"
+        case "faulted":
+            return "identity_faulted"
         case "unresolvable":
             return "identity_unresolvable"
         case _:
@@ -569,6 +591,8 @@ def _refresh_key_failure_to_mint_error(failure: _KeyResolutionFailure) -> _Bridg
             return "invalid_refresh"
         case "unavailable":
             return "identity_unavailable"
+        case "faulted":
+            return "identity_faulted"
         case "unresolvable":
             return "identity_unresolvable"
         case _:

litellm/proxy/_experimental/mcp_server/gateway_dcr_flow.py

@@ -150,11 +150,18 @@ _CLIENT_RECORD_DEBUG_KEY: Final = "gateway_dcr_client"
 _CONNECT_FLOW_DEBUG_KEY: Final = "gateway_connect_flow"
 _AUTH_CODE_DEBUG_KEY: Final = "gateway_authorization_code"
 
-ReloadUserFailure = Literal["unresolvable", "unavailable", "no_active_key"]
+ReloadUserFailure = Literal["unresolvable", "unavailable", "faulted", "no_active_key"]
 ReloadUser = Callable[[str], Awaitable[ReloadUserFailure | None]]
 """Injected live-user revalidation (the token endpoint's mirror of admission):
-``None`` means the user is active; ``unavailable`` is a retryable DB outage; anything
-else fails the grant closed."""
+``None`` means the user is active; ``unavailable`` is a retryable DB outage; ``faulted`` is
+a DB fault retrying will not clear (still 503, worded so nobody just waits); anything else
+fails the grant closed."""
+
+_DB_UNAVAILABLE_DESCRIPTION: Final = "the gateway database is unavailable; retry"
+_DB_FAULTED_DESCRIPTION: Final = (
+    "the gateway database reported a fault that is not a transient outage; "
+    "retrying will not help until the gateway deployment is repaired"
+)
 
 PROXY_API_AUDIENCE: Final[SessionAudience] = "proxy_api"
 """The audience a native client (``lite login --pkce``, a Go CLI) asks for by sending the
@@ -659,7 +666,9 @@ def _set_flow_cookie(response: Response, request: Request, handle: str, flow: _C
 def _consent_lookup_failure_response(failure: ReloadUserFailure) -> Response:
     match failure:
         case "unavailable":
-            return _oauth_error(503, "temporarily_unavailable", "the gateway database is unavailable; retry")
+            return _oauth_error(503, "temporarily_unavailable", _DB_UNAVAILABLE_DESCRIPTION)
+        case "faulted":
+            return _oauth_error(503, "temporarily_unavailable", _DB_FAULTED_DESCRIPTION)
         case "unresolvable":
             return _oauth_error(500, "server_error", "the gateway is not configured to resolve users")
         case "no_active_key":
@@ -962,7 +971,9 @@ def _reload_failure_response(failure: ReloadUserFailure) -> Response:
     ``ReloadUserFailure`` member is a type error here rather than silently 400ing."""
     match failure:
         case "unavailable":
-            return _oauth_error(503, "temporarily_unavailable", "the gateway database is unavailable; retry")
+            return _oauth_error(503, "temporarily_unavailable", _DB_UNAVAILABLE_DESCRIPTION)
+        case "faulted":
+            return _oauth_error(503, "temporarily_unavailable", _DB_FAULTED_DESCRIPTION)
         case "unresolvable":
             return _oauth_error(500, "server_error", "the gateway is not configured to resolve users")
         case "no_active_key":
@@ -981,7 +992,7 @@ def _mint_failure_response(failure: ProxyCredentialMintFailure) -> Response:
             return _oauth_error(
                 400, "invalid_grant", "this user belongs to a team; sign in again and pick the team for this credential"
             )
-        case "unavailable" | "unresolvable" | "no_active_key":
+        case "unavailable" | "faulted" | "unresolvable" | "no_active_key":
             return _reload_failure_response(failure)
         case _:
             assert_never(failure)
@@ -1297,8 +1308,8 @@ async def introspect_gateway_token(
         if peeked == "claimed":
             return _inactive_introspection_response()
     failure: Final = await reload_user(opened.principal.user_id)
-    if failure == "unavailable":
-        return _oauth_error(503, "temporarily_unavailable", "the gateway database is unavailable; retry")
+    if failure == "unavailable" or failure == "faulted":
+        return _reload_failure_response(failure)
     if failure is not None:
         return _inactive_introspection_response()
     return _active_introspection_response(opened)

litellm/proxy/_experimental/mcp_server/tool_search.py

@@ -2,20 +2,31 @@ from __future__ import annotations
 
 import json
 from collections.abc import Mapping, Sequence
+from dataclasses import dataclass
 from datetime import datetime
+from types import MappingProxyType
 from typing import TYPE_CHECKING, Any, Final, TypedDict, assert_never
 
+from pydantic import ValidationError
 from typing_extensions import ReadOnly, Required
 
 import litellm
 from litellm.proxy.agent_endpoints.agent_search import DEFAULT_AGENT_SEARCH_TOP_K
+from litellm.proxy.common_utils.semantic_text_index import (
+    Embedder,
+    EmbeddingFailed,
+    SemanticTextIndex,
+    router_embedder,
+)
+from litellm.types.mcp import MCPToolSearchSettings
 
 if TYPE_CHECKING:
-    from mcp.types import CallToolResult
+    from mcp.types import CallToolResult, Tool
 
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
     from litellm.proxy._types import UserAPIKeyAuth
 
+MCP_TOOL_SEARCH_SETTINGS_KEY: Final[str] = "mcp_tool_search"
 MCP_TOOL_SEARCH_TOOL_NAME: Final[str] = "mcp_tool_search"
 MCP_TOOL_CALL_TOOL_NAME: Final[str] = "mcp_tool_call"
 AGENT_SEARCH_TOOL_NAME: Final[str] = "agent_search"
@@ -29,17 +40,91 @@ def coerce_top_k(value: Any, default: int = 5) -> int:
         return default
 
 
-def search_tools(query: str, tools: list[dict[str, Any]], top_k: int = 5) -> list[dict[str, Any]]:
+class ToolSearchResult(TypedDict, total=False):
+    name: Required[ReadOnly[str]]
+    description: Required[ReadOnly[str]]
+    inputSchema: Required[ReadOnly[Mapping[str, object]]]
+    score: ReadOnly[float]
+
+
+@dataclass(frozen=True, slots=True)
+class SemanticToolRanker:
+    embed: Embedder
+    embedding_model: str
+    index: SemanticTextIndex
+
+
+global_mcp_tool_search_index: Final = SemanticTextIndex()
+
+
+def mcp_tool_search_settings() -> MCPToolSearchSettings | ValidationError:
+    try:
+        return MCPToolSearchSettings.model_validate(litellm.mcp_tool_search or {})
+    except ValidationError as exc:
+        return exc
+
+
+def _tool_result(tool: Tool) -> ToolSearchResult:
+    return {"name": tool.name, "description": tool.description or "", "inputSchema": tool.inputSchema}
+
+
+def _scored_result(tool: Tool, score: float) -> ToolSearchResult:
+    return {"name": tool.name, "description": tool.description or "", "inputSchema": tool.inputSchema, "score": score}
+
+
+def _tool_text(tool: Tool) -> str:
+    return "\n".join(part for part in (tool.name, tool.description or "") if part)
+
+
+def _keyword_score(query: str, tool: Tool) -> float:
+    haystack: Final = _tool_text(tool).lower()
+    return float(sum(1 for token in query.lower().split() if token in haystack))
+
+
+def _split_core_tools(tools: Sequence[Tool], core_tools: Sequence[str]) -> tuple[tuple[Tool, ...], tuple[Tool, ...]]:
+    by_name: Final = MappingProxyType({tool.name: tool for tool in tools})
+    core: Final = tuple(by_name[name] for name in dict.fromkeys(core_tools) if name in by_name)
+    rest: Final = tuple(tool for tool in tools if tool.name not in frozenset(core_tools))
+    return core, rest
+
+
+def _top_hits(
+    tools: Sequence[Tool], scores: Sequence[float], minimum: float, limit: int
+) -> tuple[tuple[float, Tool], ...]:
+    hits: Final = ((score, tool) for score, tool in zip(scores, tools, strict=True) if score >= minimum)
+    return tuple(sorted(hits, key=lambda hit: hit[0], reverse=True)[:limit])
+
+
+def search_tools(query: str, tools: Sequence[Tool], top_k: int = 5) -> tuple[ToolSearchResult, ...]:
+    """Keyword fallback used when no embedding model is configured: one point per query token found in the tool."""
     if not query:
-        return []
-    tokens: Final = query.lower().split()
+        return ()
+    scores: Final = tuple(_keyword_score(query, tool) for tool in tools)
+    return tuple(_tool_result(tool) for _, tool in _top_hits(tools, scores, minimum=1.0, limit=top_k))
 
-    def _score(tool: dict[str, Any]) -> int:
-        haystack: Final = (tool.get("name", "") + " " + tool.get("description", "")).lower()
-        return sum(1 for t in tokens if t in haystack)
 
-    scored: Final = ((s, tool) for tool in tools if (s := _score(tool)) > 0)
-    return [tool for _, tool in sorted(scored, key=lambda x: x[0], reverse=True)[:top_k]]
+async def search_mcp_tools(
+    query: str,
+    tools: Sequence[Tool],
+    top_k: int,
+    settings: MCPToolSearchSettings,
+    ranker: SemanticToolRanker | None,
+) -> tuple[ToolSearchResult, ...] | EmbeddingFailed:
+    """Core tools the caller can access come first, then up to `top_k` ranked matches from the remaining tools."""
+    core, rest = _split_core_tools(tools, settings.core_tools)
+    limit: Final = min(top_k, settings.top_k)
+    core_results: Final = tuple(_tool_result(tool) for tool in core)
+    if ranker is None:
+        return (*core_results, *search_tools(query, rest, limit))
+    if not query:
+        return core_results
+    scores: Final = await ranker.index.scores(
+        query, tuple(_tool_text(tool) for tool in rest), ranker.embed, ranker.embedding_model
+    )
+    if isinstance(scores, EmbeddingFailed):
+        return scores
+    hits: Final = _top_hits(rest, scores, minimum=settings.similarity_threshold, limit=limit)
+    return (*core_results, *(_scored_result(tool, score) for score, tool in hits))
 
 
 class _ToolParamSchema(TypedDict, total=False):
@@ -66,11 +151,17 @@ def _json_array(*items: str) -> Sequence[str]:
 
 _MCP_TOOL_SEARCH_DEFINITION: Final[VirtualToolDefinition] = {
     "name": MCP_TOOL_SEARCH_TOOL_NAME,
-    "description": "Search for MCP tools by keyword. Returns top matching tools with names, descriptions, and input schemas.",
+    "description": (
+        "Search for MCP tools by describing what you need. "
+        "Returns top matching tools with names, descriptions, and input schemas."
+    ),
     "inputSchema": {
         "type": "object",
         "properties": {
-            "query": {"type": "string", "description": "Keywords to search for in tool names and descriptions."},
+            "query": {
+                "type": "string",
+                "description": "What the tool should do, matched against names and descriptions.",
+            },
             "top_k": {"type": "integer", "description": "Maximum number of results to return.", "default": 5},
         },
         "required": _json_array("query"),
@@ -165,10 +256,28 @@ async def handle_mcp_tool_search(
     oauth2_headers: dict[str, str] | None = None,
     raw_headers: dict[str, str] | None = None,
 ) -> CallToolResult:
-    from mcp.types import CallToolResult, TextContent
-
     from litellm.proxy._experimental.mcp_server.server import _list_mcp_tools
+    from litellm.proxy.proxy_server import llm_router
 
+    settings: Final = mcp_tool_search_settings()
+    if isinstance(settings, ValidationError):
+        return _text_tool_result(
+            f"litellm_settings.{MCP_TOOL_SEARCH_SETTINGS_KEY} is invalid: {settings}", is_error=True
+        )
+    if settings.embedding_model is not None and llm_router is None:
+        return _text_tool_result(
+            f"litellm_settings.{MCP_TOOL_SEARCH_SETTINGS_KEY}.embedding_model needs a model_list so it can be called",
+            is_error=True,
+        )
+    ranker: Final = (
+        SemanticToolRanker(
+            embed=router_embedder(llm_router, settings.embedding_model, user_api_key_dict),
+            embedding_model=settings.embedding_model,
+            index=global_mcp_tool_search_index,
+        )
+        if settings.embedding_model is not None and llm_router is not None
+        else None
+    )
     mcp_listing: Final = await _list_mcp_tools(
         user_api_key_auth=user_api_key_dict,
         mcp_servers=mcp_servers,
@@ -178,17 +287,10 @@ async def handle_mcp_tool_search(
         oauth2_headers=oauth2_headers,
         raw_headers=raw_headers,
     )
-    mcp_tools: Final = mcp_listing.tools
-    tools: Final = [
-        {
-            "name": t.name,
-            "description": t.description or "",
-            "inputSchema": t.inputSchema,
-        }
-        for t in mcp_tools
-    ]
-    results: Final = search_tools(query, tools, top_k)
-    return CallToolResult(content=[TextContent(type="text", text=json.dumps(results))], isError=False)
+    results: Final = await search_mcp_tools(query, mcp_listing.tools, top_k, settings, ranker)
+    if isinstance(results, EmbeddingFailed):
+        return _text_tool_result(results.reason, is_error=True)
+    return _text_tool_result(json.dumps(results), is_error=False)
 
 
 async def handle_mcp_tool_call(

litellm/proxy/_types.py

@@ -2508,6 +2508,14 @@ class ConfigGeneralSettings(LiteLLMPydanticObjectBase):
         None,
         description="max batch input file size in MB for /v1/files uploads with purpose=batch, if a file is larger than this size it will be rejected before being forwarded to the provider",
     )
+    max_file_size_mb: int | None = Field(
+        None,
+        description="max file size in MB for /v1/files uploads, for any purpose, if a file is larger than this size it will be rejected before being forwarded to the provider",
+    )
+    blocked_file_extensions: tuple[str, ...] | None = Field(
+        None,
+        description="file extensions (e.g. ['.exe', '.sh']) rejected on /v1/files uploads, for any purpose, matched case-insensitively against the uploaded filename",
+    )
     max_response_size_mb: int | None = Field(
         None,
         description="max response size in MB, if a response is larger than this size it will be rejected",
@@ -2696,6 +2704,40 @@ class ConfigGeneralSettings(LiteLLMPydanticObjectBase):
         None,
         description="List of MCP server fields that must be filled in for a submission to pass standards checks (e.g. ['description', 'source_url', 'alias']).",
     )
+    password_policy_min_length: int | None = Field(
+        None,
+        description=(
+            "Minimum length required for a locally-managed user's password. Default is 12; "
+            "a value below 8 is floored to 8 rather than weakening the requirement further."
+        ),
+    )
+    password_policy_require_uppercase: bool | None = Field(
+        None,
+        description="If True (default), a locally-managed user's password must contain an uppercase letter.",
+    )
+    password_policy_require_lowercase: bool | None = Field(
+        None,
+        description="If True (default), a locally-managed user's password must contain a lowercase letter.",
+    )
+    password_policy_require_numbers: bool | None = Field(
+        None,
+        description="If True (default), a locally-managed user's password must contain a number.",
+    )
+    password_policy_require_special_characters: bool | None = Field(
+        None,
+        description="If True (default), a locally-managed user's password must contain a special (non-alphanumeric) character.",
+    )
+    disable_password_login_when_sso_enabled: bool | None = Field(
+        None,
+        description=(
+            "If True and SSO is configured (MICROSOFT_CLIENT_ID, GOOGLE_CLIENT_ID, "
+            "GENERIC_CLIENT_ID, or SAML_IDP_METADATA_URL/XML), disables username/password "
+            "login on /login, /v2/login, and /v3/login so SSO is the only way to reach the "
+            "Admin UI. An admin locked out of the UI can still administer the proxy over the "
+            "API with the master key; unset this setting and restart the proxy to restore "
+            "UI username/password login. Default is False."
+        ),
+    )
     disable_budget_reservation: bool | None = Field(
         None,
         description=(

litellm/proxy/agent_endpoints/agent_registry.py

@@ -6,8 +6,12 @@ from datetime import datetime, timezone
 from types import MappingProxyType
 from typing import TYPE_CHECKING, Final, NamedTuple, Protocol, TypedDict
 
+from pydantic import TypeAdapter, ValidationError
+
 import litellm
+from litellm.constants import REDACTED_BY_LITELM_STRING
 from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
+from litellm.litellm_core_utils.sensitive_data_masker import SensitiveDataMasker
 from litellm.proxy.management_helpers.object_permission_utils import (
     handle_update_object_permission_common,
 )
@@ -52,6 +56,9 @@ class AgentRecord(Protocol):
     @property
     def agent_name(self) -> str: ...
 
+    @property
+    def litellm_params(self) -> Mapping[str, object] | None: ...
+
     @property
     def object_permission_id(self) -> str | None: ...
 
@@ -121,6 +128,188 @@ def _dump_agent_params(raw: Mapping[str, object]) -> dict[str, object]:
     return dict(raw) if raw else {}
 
 
+_AGENT_PARAMS_MASKER: Final = SensitiveDataMasker()
+_REDACT_AGENT_PARAMS_MAX_DEPTH: Final = 10
+_AGENT_PARAMS_ADAPTER: Final[TypeAdapter[dict[str, object]]] = TypeAdapter(
+    dict[str, object]
+)  # mutable-ok: safe_dumps() and AgentResponse.litellm_params both require a real dict, not a Mapping
+_AGENT_PARAMS_SEQUENCE_ADAPTER: Final[TypeAdapter[tuple[object, ...]]] = TypeAdapter(tuple[object, ...])
+_EMPTY_LITELLM_PARAMS: Final[Mapping[str, object]] = MappingProxyType({})
+
+
+def redact_sensitive_agent_litellm_params(litellm_params: object, _depth: int = 0) -> object:
+    """
+    Replace credential-bearing values in an agent's litellm_params with
+    ``REDACTED_BY_LITELM_STRING`` while preserving non-secret keys (``model``,
+    ``is_public``, rate-limit config). Used so list/get/create/update
+    responses never echo a stored provider credential back to the caller.
+
+    Handles a plain dict, a JSON-serialized string (some callers hold the
+    in-memory registry's params that way), and ``None`` at the top level;
+    anything else is passed through. Recursion depth is bounded to match the
+    convention documented in ``tests/code_coverage_tests/recursive_detector.py``.
+    """
+    if litellm_params is None:
+        return None
+    if isinstance(litellm_params, str):
+        if _depth >= _REDACT_AGENT_PARAMS_MAX_DEPTH:
+            return REDACTED_BY_LITELM_STRING
+        try:
+            parsed_params: Final = _AGENT_PARAMS_ADAPTER.validate_json(litellm_params)
+        except ValidationError:
+            return REDACTED_BY_LITELM_STRING
+        return json.dumps(_redact_agent_params_tree(parsed_params, _depth + 1))
+    return _redact_agent_params_tree(litellm_params, _depth)
+
+
+def _redact_agent_params_tree(value: object, _depth: int) -> object:
+    """Structural recursion over an already-parsed litellm_params value: a
+    dict redacts sensitive keys and recurses into the rest, a list redacts
+    each element (so a secret nested inside a list of provider configs is
+    still caught), and anything else -- including a plain string leaf, which
+    must never be re-interpreted as a JSON blob -- passes through unchanged.
+    """
+    if _depth >= _REDACT_AGENT_PARAMS_MAX_DEPTH:
+        return REDACTED_BY_LITELM_STRING
+    if isinstance(value, list):
+        typed_items: Final = _AGENT_PARAMS_SEQUENCE_ADAPTER.validate_python(value)
+        return tuple(_redact_agent_params_tree(item, _depth + 1) for item in typed_items)
+    if not isinstance(value, dict):
+        return value
+    typed_params: Final = _AGENT_PARAMS_ADAPTER.validate_python(value)
+    return {
+        key: (
+            REDACTED_BY_LITELM_STRING
+            if _AGENT_PARAMS_MASKER.is_sensitive_key(key)
+            else _redact_agent_params_tree(nested_value, _depth + 1)
+        )
+        for key, nested_value in typed_params.items()
+    }  # mutable-ok: consumed by json.dumps()/AgentResponse.litellm_params, both of which require a real dict
+
+
+def parse_agent_litellm_params(value: object) -> Mapping[str, object]:
+    """Normalize a stored litellm_params column to a read-only mapping.
+
+    The prisma Json column comes back as either an already-parsed dict or a
+    JSON string depending on the read path, so handle both rather than
+    assuming one. Only ever read from (merge-source lookups), never mutated
+    or re-serialized directly, so a read-only view is enough here.
+    """
+    if isinstance(value, str):
+        try:
+            return _AGENT_PARAMS_ADAPTER.validate_json(value)
+        except ValidationError:
+            return _EMPTY_LITELLM_PARAMS
+    if isinstance(value, Mapping):
+        try:
+            return _AGENT_PARAMS_ADAPTER.validate_python(value)
+        except ValidationError:
+            return _EMPTY_LITELLM_PARAMS
+    return _EMPTY_LITELLM_PARAMS
+
+
+_MISSING_AGENT_PARAM: Final = object()
+_RESTORE_AGENT_PARAMS_MAX_DEPTH: Final = 10
+
+
+def _restore_redacted_nested_value(incoming_value: object, existing_value: object, _depth: int) -> object:
+    """Recurse into a non-sensitively-named dict/list value so a secret
+    nested underneath it (e.g. inside a list of per-provider configs) is
+    still restored, not just top-level keys. Mirrors the shapes
+    ``redact_sensitive_agent_litellm_params`` recurses into on read, so
+    restore and redact stay symmetric.
+
+    List elements are paired with the existing list by position: with no
+    stable per-element identity in an arbitrary ``dict[str, object]`` schema,
+    index is the same correspondence every other part of this restore (and
+    the endpoints' existing full-replace-on-PUT semantics) already assumes.
+    This correctly preserves a masked secret across an ordinary edit of that
+    same entry's other fields; it does not protect against a caller who both
+    reorders/resizes the list AND echoes back a masked marker in the same
+    request, which is a known, narrow limitation (see LIT-6736 PR discussion)
+    rather than a cross-entry credential leak in the common case.
+
+    A value collapsed to the flat marker by the read side's depth cap is
+    recovered wholesale from ``existing_value`` (rather than the marker
+    string itself getting persisted) whenever ``existing_value`` isn't
+    already that same flat marker. Depth-bounded like its read-side
+    counterpart; a value at the cap is returned unchanged rather than
+    corrupted.
+    """
+    if incoming_value == REDACTED_BY_LITELM_STRING and existing_value != REDACTED_BY_LITELM_STRING:
+        return existing_value
+    if _depth >= _RESTORE_AGENT_PARAMS_MAX_DEPTH:
+        return incoming_value
+    if isinstance(incoming_value, Mapping):
+        typed_incoming_map: Final = _AGENT_PARAMS_ADAPTER.validate_python(incoming_value)
+        existing_map: Final = (
+            _AGENT_PARAMS_ADAPTER.validate_python(existing_value)
+            if isinstance(existing_value, Mapping)
+            else _EMPTY_LITELLM_PARAMS
+        )
+        return _restore_redacted_litellm_params(typed_incoming_map, existing_map, _depth + 1)
+    if isinstance(incoming_value, (list, tuple)):
+        typed_incoming_seq: Final = _AGENT_PARAMS_SEQUENCE_ADAPTER.validate_python(incoming_value)
+        existing_seq: Final = (
+            _AGENT_PARAMS_SEQUENCE_ADAPTER.validate_python(existing_value)
+            if isinstance(existing_value, (list, tuple))
+            else ()
+        )
+        return tuple(
+            _restore_redacted_nested_value(
+                item,
+                existing_seq[index] if index < len(existing_seq) else None,
+                _depth + 1,
+            )
+            for index, item in enumerate(typed_incoming_seq)
+        )
+    return incoming_value
+
+
+def _resolved_agent_param_value(
+    key: str,
+    incoming: Mapping[str, object],
+    existing: Mapping[str, object],
+    _depth: int,
+) -> object:
+    """The value ``key`` should end up with in a restored litellm_params, or
+    ``_MISSING_AGENT_PARAM`` when it should be dropped entirely."""
+    if key in incoming:
+        value: Final = incoming[key]
+        if _AGENT_PARAMS_MASKER.is_sensitive_key(key):
+            return existing.get(key, _MISSING_AGENT_PARAM) if value == REDACTED_BY_LITELM_STRING else value
+        return _restore_redacted_nested_value(value, existing.get(key), _depth)
+    if _AGENT_PARAMS_MASKER.is_sensitive_key(key):
+        return existing.get(key, _MISSING_AGENT_PARAM)
+    return _MISSING_AGENT_PARAM
+
+
+def _restore_redacted_litellm_params(
+    incoming: Mapping[str, object],
+    existing: Mapping[str, object],
+    _depth: int = 0,
+) -> dict[str, object]:
+    """Restore the real credential behind any litellm_params value the caller
+    echoed back as ``REDACTED_BY_LITELM_STRING``, and behind any sensitive key
+    omitted entirely, so an edit to an unrelated field never overwrites (or
+    silently drops) a stored provider credential -- the UI never has to
+    read-and-resend a secret to keep it. Recurses into nested dicts and lists
+    so a secret nested under a non-sensitively-named key is restored too.
+
+    A sensitive key given a real (non-marker) value, including an explicit
+    empty string, is treated as a deliberate update -- that's how a caller
+    clears a credential. Non-sensitive keys always take the incoming value
+    (recursed into), matching the endpoints' existing full-replace-on-PUT /
+    merge-on-PATCH semantics for everything that isn't a secret.
+    """
+    all_keys: Final = frozenset(incoming) | frozenset(existing)
+    return {
+        key: value
+        for key in all_keys
+        if (value := _resolved_agent_param_value(key, incoming, existing, _depth)) is not _MISSING_AGENT_PARAM
+    }  # mutable-ok: fed to safe_dumps() for JSON-column storage, which requires a real dict
+
+
 class GrantMigrationResult(NamedTuple):
     rewritten: int
     missed: int
@@ -301,9 +490,14 @@ class AgentRegistry:
         try:
             agent_name: Final = agent.get("agent_name")
 
-            # Serialize litellm_params
+            # Serialize litellm_params. A create has no stored row to restore a
+            # secret behind, so a sensitive key submitted as the redaction
+            # marker (e.g. a stray client re-post) is dropped rather than
+            # persisted as the literal placeholder string.
             litellm_params_obj: Final = agent.get("litellm_params", {})
-            litellm_params_dict: Final[dict[str, object]] = _dump_agent_params(litellm_params_obj)
+            litellm_params_dict: Final = _restore_redacted_litellm_params(
+                _dump_agent_params(litellm_params_obj), _EMPTY_LITELLM_PARAMS
+            )
             litellm_params: Final[str] = safe_dumps(litellm_params_dict)
 
             # Serialize agent_card_params
@@ -410,8 +604,14 @@ class AgentRegistry:
             update_data: Final[dict[str, object]] = {}
             if augment_agent.get("agent_name"):
                 update_data["agent_name"] = augment_agent.get("agent_name")
-            if augment_agent.get("litellm_params"):
-                update_data["litellm_params"] = safe_dumps(augment_agent.get("litellm_params"))
+            if "litellm_params" in agent:
+                existing_litellm_params: Final = parse_agent_litellm_params(existing_agent.get("litellm_params"))
+                update_data["litellm_params"] = safe_dumps(
+                    _restore_redacted_litellm_params(
+                        _dump_agent_params(agent.get("litellm_params") or _EMPTY_LITELLM_PARAMS),
+                        existing_litellm_params,
+                    )
+                )
             if augment_agent.get("agent_card_params"):
                 update_data["agent_card_params"] = safe_dumps(augment_agent.get("agent_card_params"))
 
@@ -474,9 +674,22 @@ class AgentRegistry:
         try:
             agent_name: Final = agent.get("agent_name")
 
+            # A PUT fully replaces litellm_params from the request body, so the
+            # existing row is read up front to restore any sensitive key the
+            # caller echoed back redacted (or omitted) rather than persisting
+            # the marker -- or nothing -- over the real stored credential.
+            existing_row: Final = await agents_table(prisma_client).find_unique(
+                where={"agent_id": agent_id}  # mutable-ok: prisma's query builder rejects a Mapping/MappingProxyType
+            )
+            existing_litellm_params: Final = parse_agent_litellm_params(
+                existing_row.litellm_params if existing_row is not None else None
+            )
+
             # Serialize litellm_params
             litellm_params_obj: Final = agent.get("litellm_params", {})
-            litellm_params_dict: Final[dict[str, object]] = _dump_agent_params(litellm_params_obj)
+            litellm_params_dict: Final = _restore_redacted_litellm_params(
+                _dump_agent_params(litellm_params_obj), existing_litellm_params
+            )
             litellm_params: Final[str] = safe_dumps(litellm_params_dict)
 
             # Serialize agent_card_params
@@ -512,9 +725,8 @@ class AgentRegistry:
                     update_data[rate_field] = _val
 
             if agent.get("object_permission") is not None:
-                existing_agent: Final = await agents_table(prisma_client).find_unique(where={"agent_id": agent_id})
                 existing_object_permission_id: Final = (
-                    existing_agent.object_permission_id if existing_agent is not None else None
+                    existing_row.object_permission_id if existing_row is not None else None
                 )
                 agent_copy: Final = dict(agent)
                 object_permission_id: Final = await handle_update_object_permission_common(

litellm/proxy/agent_endpoints/agent_search.py

@@ -2,17 +2,18 @@
 
 from __future__ import annotations
 
-import math
-from collections.abc import Awaitable, Mapping, Sequence
+from collections.abc import Sequence
 from dataclasses import dataclass
-from itertools import chain
-from types import MappingProxyType
-from typing import TYPE_CHECKING, Final, Protocol, TypeAlias
+from typing import TYPE_CHECKING, Final, TypeAlias
 
-from openai import OpenAIError
 from pydantic import BaseModel, ConfigDict, ValidationError
 
-from litellm.exceptions import BudgetExceededError
+from litellm.proxy.common_utils.semantic_text_index import (
+    Embedder,
+    EmbeddingFailed,
+    SemanticTextIndex,
+    router_embedder,
+)
 from litellm.types.agents import AgentResponse
 
 if TYPE_CHECKING:
@@ -21,12 +22,6 @@ if TYPE_CHECKING:
 
 DEFAULT_AGENT_SEARCH_TOP_K: Final = 5
 
-Vector: TypeAlias = tuple[float, ...]
-
-
-class Embedder(Protocol):
-    def __call__(self, texts: Sequence[str]) -> Awaitable[Sequence[Vector]]: ...
-
 
 @dataclass(frozen=True, slots=True)
 class AgentSearchHit:
@@ -67,18 +62,6 @@ class _SearchableCard(BaseModel):
     skills: tuple[_SearchableSkill, ...] = ()
 
 
-class _EmbeddingItem(BaseModel):
-    model_config = ConfigDict(frozen=True, extra="ignore")
-
-    embedding: tuple[float, ...]
-
-
-class _EmbeddingData(BaseModel):
-    model_config = ConfigDict(frozen=True, extra="ignore")
-
-    data: tuple[_EmbeddingItem, ...]
-
-
 class AgentSearchResult(BaseModel):
     model_config = ConfigDict(frozen=True)
 
@@ -117,110 +100,21 @@ def agent_search_result(hit: AgentSearchHit) -> AgentSearchResult:
     )
 
 
-def cosine_similarity(left: Vector, right: Vector) -> float:
-    dot: Final = sum(a * b for a, b in zip(left, right, strict=True))
-    norms: Final = math.sqrt(sum(a * a for a in left)) * math.sqrt(sum(b * b for b in right))
-    return dot / norms if norms else 0.0
-
-
-def embedding_spend_metadata(user_api_key_dict: UserAPIKeyAuth) -> dict[str, object]:
-    from litellm.proxy.litellm_pre_call_utils import LiteLLMProxyRequestSetup
-
-    return {  # mutable-ok: the router mutates the metadata dict it is handed
-        **LiteLLMProxyRequestSetup.get_sanitized_user_information_from_key(user_api_key_dict),
-        "user_api_key": user_api_key_dict.api_key,
-    }
-
-
-def router_embedder(router: Router, embedding_model: str, user_api_key_dict: UserAPIKeyAuth) -> Embedder:
-    async def embed(texts: Sequence[str]) -> Sequence[Vector]:
-        batch: Final = list(texts)  # mutable-ok: Router.aembedding accepts only str | list input
-        response: Final = await router.aembedding(
-            model=embedding_model, input=batch, metadata=embedding_spend_metadata(user_api_key_dict)
-        )
-        return tuple(item.embedding for item in _EmbeddingData.model_validate(response.model_dump()).data)
-
-    return embed
-
-
-_NO_VECTORS: Final[Mapping[str, Vector]] = MappingProxyType({})
-
-
-async def _embed_all(embed: Embedder, texts: Sequence[str]) -> tuple[Vector, ...] | AgentSearchEmbeddingFailed:
-    try:
-        vectors: Final = tuple(await embed(texts))
-    except (OpenAIError, ValueError, BudgetExceededError) as exc:
-        return AgentSearchEmbeddingFailed(reason=f"embedding the search query failed: {exc}")
-    if len(vectors) != len(texts):
-        return AgentSearchEmbeddingFailed(
-            reason=f"embedding model returned {len(vectors)} vectors for {len(texts)} inputs"
-        )
-    return vectors
-
-
-@dataclass(frozen=True, slots=True)
-class _Embedded:
-    query_vector: Vector
-    vectors: Mapping[str, Vector]
-
-
-def _same_dimension(query_vector: Vector, vectors: Mapping[str, Vector], texts: Sequence[str]) -> bool:
-    return all(len(vectors[text]) == len(query_vector) for text in texts)
-
-
-async def _embed_query_and_agents(
-    embed: Embedder, query: str, texts: Sequence[str], cached: Mapping[str, Vector]
-) -> _Embedded | AgentSearchEmbeddingFailed:
-    missing: Final = tuple(dict.fromkeys(text for text in texts if text not in cached))
-    embedded: Final = await _embed_all(embed, (query, *missing))
-    if isinstance(embedded, AgentSearchEmbeddingFailed):
-        return embedded
-    vectors: Final = MappingProxyType(dict(chain(cached.items(), zip(missing, embedded[1:], strict=True))))
-    if _same_dimension(embedded[0], vectors, texts):
-        return _Embedded(query_vector=embedded[0], vectors=vectors)
-    unique: Final = tuple(dict.fromkeys(texts))
-    reembedded: Final = await _embed_all(embed, (query, *unique))
-    if isinstance(reembedded, AgentSearchEmbeddingFailed):
-        return reembedded
-    return _Embedded(
-        query_vector=reembedded[0], vectors=MappingProxyType(dict(zip(unique, reembedded[1:], strict=True)))
-    )
-
-
 class AgentSearchIndex:
     """Caches one vector per distinct agent text per embedding model, so repeat searches only embed the query."""
 
     def __init__(self) -> None:
-        self._vectors: Mapping[str, Mapping[str, Vector]] = MappingProxyType({})
-
-    def _merged(self, embedding_model: str, embedded: _Embedded) -> Mapping[str, Vector]:
-        kept: Final = {
-            text: vector
-            for text, vector in self._vectors.get(embedding_model, _NO_VECTORS).items()
-            if len(vector) == len(embedded.query_vector)
-        }
-        return MappingProxyType({**kept, **embedded.vectors})
+        self._index: Final = SemanticTextIndex()
 
     async def search(
         self, query: str, agents: Sequence[AgentResponse], top_k: int, embed: Embedder, embedding_model: str
     ) -> AgentSearchHits | AgentSearchEmbeddingFailed:
-        if not agents:
-            return AgentSearchHits(hits=())
         texts: Final = tuple(agent_search_text(agent) for agent in agents)
-        cached: Final = self._vectors.get(embedding_model, _NO_VECTORS)
-        embedded: Final = await _embed_query_and_agents(embed, query, texts, cached)
-        if isinstance(embedded, AgentSearchEmbeddingFailed):
-            return embedded
-        if not _same_dimension(embedded.query_vector, embedded.vectors, texts):
-            return AgentSearchEmbeddingFailed(
-                reason=f"embedding model {embedding_model} returned vectors of mixed dimensions"
-            )
-        self._vectors = MappingProxyType({**self._vectors, embedding_model: self._merged(embedding_model, embedded)})
+        scores: Final = await self._index.scores(query, texts, embed, embedding_model)
+        if isinstance(scores, EmbeddingFailed):
+            return AgentSearchEmbeddingFailed(reason=scores.reason)
         ranked: Final = sorted(
-            (
-                AgentSearchHit(agent=agent, score=cosine_similarity(embedded.query_vector, embedded.vectors[text]))
-                for agent, text in zip(agents, texts, strict=True)
-            ),
+            (AgentSearchHit(agent=agent, score=score) for agent, score in zip(agents, scores, strict=True)),
             key=lambda hit: hit.score,
             reverse=True,
         )

litellm/proxy/agent_endpoints/endpoints.py

@@ -20,7 +20,6 @@ from typing_extensions import ReadOnly, Required
 
 import litellm
 from litellm._logging import verbose_proxy_logger
-from litellm.litellm_core_utils.litellm_logging import _get_masked_values
 from litellm.llms.custom_httpx.http_handler import get_async_httpx_client
 from litellm.proxy._types import (
     CommonProxyErrors,
@@ -33,6 +32,10 @@ from litellm.proxy.a2a.agent_card import (
     merge_agent_card,
     normalize_protocol_version,
 )
+from litellm.proxy.agent_endpoints.agent_registry import (
+    parse_agent_litellm_params,
+    redact_sensitive_agent_litellm_params,
+)
 from litellm.proxy.agent_endpoints.agent_search import (
     DEFAULT_AGENT_SEARCH_TOP_K,
     AgentSearchEmbeddingFailed,
@@ -139,25 +142,37 @@ async def _attach_keys_to_agents(agents: Sequence[AgentResponse], prisma_client)
         agent.keys = matched_keys or None
 
 
+def _redact_agent_litellm_params_dict(
+    litellm_params: Mapping[str, object],
+) -> dict[str, object]:  # mutable-ok: AgentResponse.litellm_params is declared as a plain dict, not Mapping
+    """Type-narrowing wrapper: a dict in always yields a dict back from
+    ``redact_sensitive_agent_litellm_params``, which the function's general
+    (possible-JSON-string, possibly-None) signature can't express."""
+    return dict(  # mutable-ok: AgentResponse.litellm_params is declared as a plain dict, not Mapping
+        parse_agent_litellm_params(redact_sensitive_agent_litellm_params(litellm_params))
+    )
+
+
 def _redact_sensitive_agent_fields(
     agents: Sequence[AgentResponse],
+    *,
+    is_admin: bool,
 ) -> list[AgentResponse]:
     """
-    Return copies of the given agents with sensitive configuration fields
-    redacted.  The original objects are not modified.
+    Return copies of the given agents with credential-bearing litellm_params
+    values replaced by a fixed marker (never returned to ANY caller,
+    admin included) and, for non-admin callers, virtual-key and header
+    fields stripped entirely. The original objects are not modified.
     """
     redacted: Final[list[AgentResponse]] = []
     for agent in agents:
         copy = agent.model_copy(deep=True)
-        copy.static_headers = None
-        copy.extra_headers = None
-        copy.keys = None
+        if not is_admin:
+            copy.static_headers = None
+            copy.extra_headers = None
+            copy.keys = None
         if copy.litellm_params:
-            copy.litellm_params = _get_masked_values(
-                copy.litellm_params,
-                unmasked_length=4,
-                number_of_asterisks=4,
-            )
+            copy.litellm_params = _redact_agent_litellm_params_dict(copy.litellm_params)
         redacted.append(copy)
     return redacted
 
@@ -345,13 +360,13 @@ async def get_agents(
                 global_agent_registry.ids_for_agent(agent.agent_id).isdisjoint(litellm.public_agent_groups)
             )
 
-        # Redact sensitive fields for non-admin users
+        # litellm_params secrets are always redacted; keys/headers stay
+        # admin-only.
         is_admin: Final = (
             user_api_key_dict.user_role == LitellmUserRoles.PROXY_ADMIN
             or user_api_key_dict.user_role == LitellmUserRoles.PROXY_ADMIN.value
         )
-        if not is_admin:
-            returned_agents = _redact_sensitive_agent_fields(returned_agents)
+        returned_agents = _redact_sensitive_agent_fields(returned_agents, is_admin=is_admin)
 
         if health_check:
             agents_with_url: Final = [agent for agent in returned_agents if (agent.agent_card_params or {}).get("url")]
@@ -505,7 +520,9 @@ async def create_agent(
                 "Failed to register agent '%s' (ID: %s) in memory: %s", agent_name, agent_id, reg_error
             )
 
-        return result
+        # The caller is a proxy admin (enforced above); litellm_params
+        # secrets are still never echoed back in the response.
+        return _redact_sensitive_agent_fields((result,), is_admin=True)[0]
 
     except HTTPException:
         raise
@@ -578,13 +595,13 @@ async def get_agent_by_id(
 
         await _attach_keys_to_agents([agent], prisma_client)
 
-        # Redact sensitive fields for non-admin users
+        # litellm_params secrets are always redacted; keys/headers stay
+        # admin-only.
         is_admin = (
             user_api_key_dict.user_role == LitellmUserRoles.PROXY_ADMIN
             or user_api_key_dict.user_role == LitellmUserRoles.PROXY_ADMIN.value
         )
-        if not is_admin:
-            agent = _redact_sensitive_agent_fields([agent])[0]
+        agent = _redact_sensitive_agent_fields((agent,), is_admin=is_admin)[0]
 
         return agent
     except HTTPException:
@@ -688,7 +705,7 @@ async def update_agent(
             "Successfully updated agent '%s' (ID: %s) in memory", existing_agent.get("agent_name"), agent_id
         )
 
-        return result
+        return _redact_sensitive_agent_fields((result,), is_admin=True)[0]
     except HTTPException:
         raise
     except Exception as e:
@@ -791,7 +808,7 @@ async def patch_agent(
             "Successfully updated agent '%s' (ID: %s) in memory", existing_agent.get("agent_name"), agent_id
         )
 
-        return result
+        return _redact_sensitive_agent_fields((result,), is_admin=True)[0]
     except HTTPException:
         raise
     except Exception as e: