The Gemini REST API documents the embedding task type parameter as
camelCase `taskType`. The existing transformation functions convert
`dimensions` to `outputDimensionality` but miss the parallel
`task_type` to `taskType` conversion. This adds that conversion to
both `transform_openai_input_gemini_content` (batchEmbedContents path)
and `transform_openai_input_gemini_embed_content` (embedContent path).
Fixes#24190
Guard os.waitpid and os.WNOHANG usage with sys.platform check.
These APIs are Unix-only; on Windows they cause AttributeError
and prevent proxy startup.
- _try_waitpid_watch: return False on Windows, fall back to
os.kill polling
- _reap_all_zombies: return empty set on Windows (no zombies)
Add unit tests for Windows path.
Made-with: Cursor
- rbac_utils.py: change feature_name from str to Literal["agents", "vector_stores"]
so typos are caught by type checkers at import time
- proxy_setting_endpoints.py: extract _RUNTIME_GENERAL_SETTINGS_FLAGS as a module-level
constant, replacing duplicated inline lists in get_ui_settings and update_ui_settings
- test_vector_store_rbac.py: remove try/except pattern that silently swallowed non-403
HTTPExceptions; tests now let any unexpected exception propagate as a test failure
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- rbac_utils.py: remove duplicated _check_if_team_admin/_is_user_team_admin_for_any_team;
delegate to _user_has_admin_privileges from management_endpoints/common_utils with the
shared user_api_key_cache (fixes no-op DualCache and missing org admin coverage)
- test_rbac_utils.py: update patch target to match new delegation path
- SidebarProvider.tsx: pass allowAgentsForTeamAdmins and allowVectorStoresForTeamAdmins
props to Sidebar
- leftnav.tsx: add useTeams hook + isTeamAdmin memo; exempt team admins from sidebar
filtering when allow_*_for_team_admins is enabled (fixes frontend/backend inconsistency)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add proxy-admin-configurable toggles to restrict internal users (and optionally
team admins) from accessing agent and vector store management features.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Verifies that vertex_ai gemini models route to
aiplatform.googleapis.com instead of
generativelanguage.googleapis.com, preventing
regressions if the branch ordering changes.
When create_batch routes via x-litellm-model header, the response batch_id
was returned raw without model routing info. This meant retrieve_batch could
not determine which provider/credentials to use, defaulting to "openai"
instead of the correct provider (e.g., VLLM).
Now encodes batch_id, output_file_id, and error_file_id with model info
(same pattern as the model-embedded file_id flow in Scenario 1), so
retrieve_batch can decode and route back to the correct provider.
When the Prisma query engine process is alive but not accepting
connections (e.g., startup race condition in containerized
deployments), lightweight reconnects (disconnect + connect) will
never succeed. The health watchdog retries indefinitely without
escalating to a full Prisma client recreation.
Adds a consecutive failure counter that triggers a heavy reconnect
(full Prisma client and engine recreation) after 3 consecutive
lightweight reconnect failures (configurable via
PRISMA_RECONNECT_ESCALATION_THRESHOLD env var).
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: add missing LiteLLM_ClaudeCodePluginTable to schema.prisma
- Claude Code Plugin Marketplace endpoints (/claude-code/marketplace.json,
/claude-code/plugins) were returning 500 errors because
LiteLLM_ClaudeCodePluginTable model was missing from both schema.prisma files
- Prisma client was generated without this table causing AttributeError:
'Prisma' object has no attribute 'litellm_claudecodeplugintable'
- Added missing model definition to root schema.prisma and
litellm/proxy/schema.prisma
Fixes#21310
* test: add regression test for LiteLLM_ClaudeCodePluginTable schema
* fix: address greptile review - add @updatedAt, clean up test imports
- Add unit test that scans Python source for Base64 Basic Auth patterns
that would be flagged by secret scanners like GitGuardian/ggshield
- Add secret-scan job to the linting CI workflow that runs the test on
every PR and optionally runs ggshield if GITGUARDIAN_API_KEY is set
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ModelResponse.choices was typed as List[Union[Choices, StreamingChoices]] which
caused Pydantic serialization warnings and false linting errors. Now that
ModelResponseStream exists for streaming, narrow ModelResponse.choices to
List[Choices] and migrate all ModelResponse(stream=True) call sites to use
ModelResponseStream() instead.
Vertex AI batch IDs are plain numeric strings (e.g., "3814889423749775360")
unlike OpenAI's "batch_"-prefixed IDs. encode_file_id_with_model() was
defaulting to "file-" prefix for unrecognized ID formats, causing Vertex AI
batch responses to return IDs like "file-bGl0ZWxsbTox..." instead of the
expected "batch_..." prefix per the OpenAI Batch API contract.
Add an optional id_type parameter to encode_file_id_with_model() so the
batch creation endpoint can specify id_type="batch" when encoding batch
response IDs. Default remains "file" for backward compatibility.
Closes#18192
Move the fix to the OpenRouter level: define native OpenRouter models
(openrouter/auto, openrouter/free, openrouter/bodybuilder) and check
them in get_llm_provider() before the provider_list stripping logic.
This prevents the second strip across all bridges without modifying
each adapter/handler individually.
Fixes#16353
Add routing prefixes bedrock/nova/<ARN> and bedrock/nova-2/<ARN> so
LiteLLM can identify the base model family for custom/imported Nova
models and enable the correct supported params (tools, web_search,
reasoning_effort).
Changes:
- Route nova/ and nova-2/ prefixed models to converse API
- Strip spec prefix before sending ARN to Bedrock
- Return sentinel base models (amazon.nova-custom, amazon.nova-2-custom)
so downstream Nova checks work
- Recognize nova-2/ prefix in _is_nova_2_model() for reasoning support
- Handle nova/nova-2 in get_bedrock_model_id() for proper ARN encoding
- Add unit tests for all new behavior
Allow passing enable_json_schema_validation as a parameter to completion()
and acompletion() instead of only relying on the global
litellm.enable_json_schema_validation flag. The per-request value takes
priority when provided; otherwise falls back to the global (backward
compatible). This makes JSON schema validation safe for concurrent usage
in FastAPI and other multi-threaded environments.
* fix(anthropic): filter unsupported JSON schema constraints for structured outputs
Fixes 400 error when using Anthropic models with structured outputs that have
min/max constraints.
The Anthropic API doesn't support these JSON schema constraints:
- minimum/maximum (numeric)
- exclusiveMinimum/exclusiveMaximum (numeric)
- minLength/maxLength (string)
- minItems/maxItems (array)
This mirrors the transformation done by the official Anthropic Python SDK.
See: https://platform.claude.com/docs/en/build-with-claude/structured-outputs#how-sdk-transformation-works
Adds tests for the schema filtering function.
* fix: update descriptions with removed constraint info in filter_anthropic_output_schema
Address review feedback: the function now appends removed constraint
information to the description field (matching Anthropic SDK behavior),
rather than silently dropping constraints.
---------
Co-authored-by: OpenClaw <openclaw@users.noreply.github.com>
* feat(bedrock): add DeepSeek V3.2 pricing and region support
* feat(bedrock): add minimax.minimax-m2.1 pricing and region support
* feat(bedrock): add moonshotai.kimi-k2.5 pricing and region support
* feat(bedrock): add qwen.qwen3-coder-next
pricing and region support
* add some sanity unit tests for the bedrock beta models added
* --amend
* resolve greptileai comments and suggestions
* fix(batch_completion): submit all model futures before waiting
* test: add batch_completion all responses concurrency regression
* fix(batch_completion): continue collecting responses on per-model failures
* fix(batch_completion): handle empty and string models in all responses
* test(batch_completion): avoid blocking wait in concurrency regression
* fix: reasoning_effort=None returns None for Opus 4.6
Previously, _map_reasoning_effort would return adaptive thinking
for Opus 4.6 even when reasoning_effort was None, which breaks the
expected contract where None means no thinking is sent.
* fix: handle reasoning_effort="none" string for Opus 4.6
The string "none" is a valid OpenAI reasoning_effort value meaning
"disable thinking". Previously it was mapped to adaptive for Opus 4.6.
* Update tests/litellm/llms/anthropic/test_anthropic_reasoning_effort.py
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
---------
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Fixes#20534
The UI sends the full form on every team update, including premium
metadata fields like `policies: []` and `team_member_key_duration: ""`.
The backend's `_update_metadata_fields` treated any non-None value as
premium feature usage and returned 403 for non-enterprise users — even
when the fields were empty and the user was just updating basic settings
like team name or budget.
Added `_has_non_empty_value` helper and use it in the premium field gate
in `_update_metadata_fields` so empty lists, blank strings, and None
values skip the premium check entirely. Non-empty values still enforce
the enterprise requirement as before.
Adds litellm.proxy_auth to automatically obtain and refresh OAuth2/JWT
tokens when connecting to LiteLLM Proxy or any OAuth2-protected endpoint.
- Add ProxyAuthHandler for token lifecycle (obtain, cache, refresh)
- Add AzureADCredential wrapper for azure-identity credentials
- Add GenericOAuth2Credential for any OAuth2 provider (Okta, Auth0, etc)
- Auto-inject Authorization headers in completion() and embedding()
Closes#19834
Fixes#19478
The stream_chunk_builder function was not handling image chunks from
models like gemini-2.5-flash-image. When streaming responses were
reconstructed (e.g., for caching), images in delta.images were lost.
This adds handling for image_chunks similar to how audio, annotations,
and other delta fields are handled.
