build(ai-gateway): build the sibling wheels from the repo so the image never waits on PyPI

litellm-rust/crates/ai-gateway/Dockerfile

@@ -51,10 +51,15 @@ RUN cargo build --locked --release -p litellm-ai-gateway --bin litellm-ai-gatewa
 # The root pyproject builds with maturin against litellm-rust/crates/python-bridge,
 # so the wheel is built here, next to the crate sources and the cargo toolchain,
 # and the runtime stage installs the artifact instead of compiling anything.
+# litellm[proxy] pins litellm-enterprise and litellm-proxy-extras to the versions
+# in this repo, and those hit PyPI hours after every version bump merges, so both
+# wheels are built from the repo too instead of being resolved from PyPI.
 COPY pyproject.toml README.md LICENSE /build/
 COPY litellm/ /build/litellm/
 COPY enterprise/ /build/enterprise/
-RUN pip3 wheel --no-cache-dir --no-deps --wheel-dir /build/dist /build
+COPY litellm-proxy-extras/ /build/litellm-proxy-extras/
+RUN pip3 wheel --no-cache-dir --no-deps --wheel-dir /build/dist \
+        /build /build/enterprise /build/litellm-proxy-extras
 
 # ---- Runtime ----------------------------------------------------------------
 # python:3.11-slim-bookworm ships libpython3.11, matching the builder's PyO3
@@ -69,10 +74,15 @@ RUN apt-get update \
 WORKDIR /app
 
 # Install litellm (with proxy extras) FROM THIS REPO'S SOURCE so
-# `import litellm.proxy.read_model_list` works — it is not on PyPI yet.
+# `import litellm.proxy.read_model_list` works — it is not on PyPI yet. The two
+# sibling wheels come from the builder as well, so the pins in litellm[proxy]
+# resolve against them and never wait on a PyPI publish.
 COPY --from=builder /build/dist/*.whl /tmp/wheels/
 RUN wheel="$(ls /tmp/wheels/litellm-*.whl)" \
-    && pip install --no-cache-dir "${wheel}[proxy]" \
+    && pip install --no-cache-dir \
+        /tmp/wheels/litellm_enterprise-*.whl \
+        /tmp/wheels/litellm_proxy_extras-*.whl \
+        "${wheel}[proxy]" \
     && rm -rf /tmp/wheels
 
 # The compiled gateway binary (pure-Rust realtime hot path; Python is load-time

litellm-rust/crates/ai-gateway/Dockerfile.dockerignore

@@ -10,6 +10,7 @@
 #   - litellm/            (pip install . needs the full package + proxy reader)
 #   - litellm-rust/       (the rust workspace; Cargo.lock + crate sources)
 #   - enterprise/         (litellm/proxy/enterprise symlinks into it; maturin walks it)
+#   - litellm-proxy-extras/ (built into a wheel alongside enterprise/ for litellm[proxy])
 #   - pyproject.toml / README.md / LICENSE  (packaging metadata for the wheel build)
 #   - rust-toolchain.toml (the pinned channel every cargo call in the build uses)
 *
@@ -18,6 +19,7 @@
 !litellm/
 !litellm-rust/
 !enterprise/
+!litellm-proxy-extras/
 !pyproject.toml
 !rust-toolchain.toml
 !README.md
@@ -28,6 +30,7 @@
 **/target/
 # Committed python distribution artifacts; the wheel build does not read them.
 enterprise/dist/
+litellm-proxy-extras/dist/
 # Python caches and compiled bytecode.
 **/__pycache__/
 **/*.pyc