feat(rust-bridge): extend native dispatch foundation to chat completions, responses, and messages

[devin-ai-integration]

TLDR

Problem this solves:

• Chat completions and Responses had no explicit catalog rule, only the implicit Python default
• Their Python bindings named native completion/acompletion/responses/aresponses that the extension never exported
• Chat completions and Messages dispatch boundaries had no behavioral tests

How it solves it:

• Explicit PYTHON_ONLY catalog rules for CHAT_COMPLETIONS and RESPONSES, matching embeddings and Messages
• Declining native completion/acompletion/responses/aresponses scaffolds, same shape as embeddings (feat(embeddings): add native dispatch foundation #42799)
• _native.pyi declarations for the four new exports
• Dispatch tests for chat completions and Messages, modeled on the embeddings tests
• Fixes the stubtest failure rust-wheel has had on main since feat(embeddings): add native dispatch foundation #42799: runtime parameter names and __all__ now match _native.pyi

Routes left Python-only after this PR: chat completions, Responses, Messages, embeddings. None of them ships a native rollout. The existing keyword-style chat_completions/achat_completions native functions and the Responses WebSocket connection are untouched. No provider-native execution is added

User Flow

Before: a developer calls any of the three routes and gets the Python result

1. They send POST https://litellm-domain/v1/chat/completions, POST https://litellm-domain/v1/responses, or POST https://litellm-domain/v1/messages
2. The Python implementation handles the request and the response comes back unchanged

After: the same calls return the same Python results

1. They send the same three requests
2. The Python implementation still handles every request, now under explicit Python-only catalog rules; nothing visible changes

Relevant issues

Follows #42799

Affected release

Linear ticket

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

This PR is foundation only: no user-visible behavior changes and no provider calls are added, so there is no live before/after to capture. Validation at the tip (b1fa7dd), merge base dc14e76:

Before (dc14e76)

1. python -c "from litellm.rust_bridge import _native; _native.completion" fails with AttributeError: module has no attribute 'completion' (same for acompletion, responses, aresponses)

After (b1fa7dd)

1. python -I -m mypy.stubtest --mypy-config-file tests/test_litellm/rust_bridge/stubtest.ini litellm.rust_bridge._native on a fresh maturin develop build: Success: no issues found in 1 module (main fails with 19 errors on the embeddings exports)
2. maturin develop then python -c "from litellm.rust_bridge import _native; _native.completion({}, (), {})" raises RustBridgeDeclined: native chat completions route is not implemented (same shape for the other three)
3. LITELLM_LOCAL_MODEL_COST_MAP=True uv run --no-sync pytest tests/test_litellm/chat_completions tests/test_litellm/messages tests/test_litellm/rust_bridge tests/test_litellm/responses/test_dispatch.py tests/test_litellm/embeddings: all pass except one preexisting tenacity import error in test_responses_with_retries_uses_the_dispatch_entrypoint, unrelated to this change
4. cargo fmt --all --check, cargo clippy -p litellm-python-bridge --all-targets, cargo test -p litellm-python-bridge routes:: (39 passed) clean
5. ruff format/ruff check and basedpyright clean on every changed Python file

Type

🆕 New Feature
✅ Test

Caveats (if any)

Low

• Native scaffolds only decline; a real route machine is follow-up work
• Catalog rules restate the existing Python default, so runtime behavior is identical

Link to Devin session: https://app.devin.ai/sessions/f5e1e146960e47a997e39f10827bd261
Open in Devin Desktop: https://app.devin.ai/desktop/session/f5e1e146960e47a997e39f10827bd261?variant=devin
Requested by: @yujonglee-berri

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Confidence Score: 4/5

The dispatch behavior appears sound, but the required fallback coverage must be completed before merging

Findings

1.  Fallback Coverage Is Incomplete ▶

Summary

This PR explicitly keeps chat completions and Responses on Python, exports declining native scaffolds for both routes, updates extension typing, and adds chat-completions and Messages dispatch coverage

• Registers completion, acompletion, responses, and aresponses in the PyO3 module
• Adds matching request and response declarations to the native stub
• Adds explicit PYTHON_ONLY catalog entries without changing current route selection
• Adds focused dispatch and argument-projection tests
• Still needs the repository-required module-missing fallback coverage for the newly exposed routes

_{Reviews (1) · Last reviewed commit: "refactor(rust-bridge): keep secret manag..."}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_native_dispatch_chat_responses_messages (b1fa7dd) with main (49d0ece)}