Litellm oss staging 04 14 2026 p1

[krrish-berri-2]

Relevant issues

Pre-Submission checklist

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

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Delays in PR merge?

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

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Screenshots / Proof of Fix

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Apr 15, 2026 3:31am

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing litellm_oss_staging_04_14_2026_p1 (20758e6) with main (9790a46)}

[greptile-apps]

Greptile Summary

This staging PR bundles four independent fixes/features: (1) Bedrock Converse now skips synthetic tool injection for json_object requests without a schema and corrects finish_reason to "stop" after filtering all synthetic tool calls; (2) litellm_metadata is now populated on the pre_call guardrail path to match the post_call path; (3) api_provider label is added to litellm_requests_metric and litellm_spend_metric; (4) a NO_OPENAPI env var disables /openapi.json, and pricing is added for openrouter/google/gemini-3.1-flash-lite-preview. All remaining findings are P2 style cleanup.

Confidence Score: 5/5

Safe to merge — all functional changes are correct and well-tested; only P2 style issues remain.

All four bundled changes are logically sound and backed by new unit tests. The three findings are P2 style issues (duplicate import, inline import, missing blank line) that do not affect runtime behavior.

litellm/proxy/proxy_server.py (duplicate import) and litellm/proxy/guardrails/guardrail_hooks/unified_guardrail/unified_guardrail.py (inline import) have minor style cleanup needed.

Important Files Changed

Filename	Overview
litellm/llms/bedrock/chat/converse_transformation.py	Skips synthetic tool injection for json_object with no schema, and fixes finish_reason to "stop" when json_mode filters all tool calls; logic is correct and well-covered by new unit tests.
litellm/proxy/proxy_server.py	Wires _get_openapi_url() into FastAPI app constructor; duplicate import of _get_openapi_url on lines 496 and 499 should be cleaned up.
litellm/proxy/utils.py	Adds _get_openapi_url() helper following the same pattern as _get_docs_url; missing one blank line before the function definition.
litellm/proxy/guardrails/guardrail_hooks/unified_guardrail/unified_guardrail.py	Adds _ensure_litellm_metadata to align pre_call guardrail with post_call; deferred import inside the function body should be moved to module level per project style guide.
litellm/types/integrations/prometheus.py	Adds api_provider label to litellm_requests_metric and litellm_spend_metric; backwards-compatible addition with appropriate tests.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["response_format param"] --> B{type?}
    B -->|json_schema + model supports native| C["outputConfig path\n(no tool injection)"]
    B -->|json_schema + fallback model| D["Inject synthetic\njson_tool_call tool"]
    B -->|json_object with schema| D
    B -->|json_object NO schema| E["Skip tool injection\n(NEW: prevents empty {} response)"]
    C --> F["finish_reason from stopReason"]
    D --> G["_filter_json_mode_tools"]
    G --> H{filtered_tools empty\nand tools had calls?}
    H -->|yes| I["finish_reason = 'stop'\n(NEW: was 'tool_calls')"]
    H -->|no| F
    E --> F

Loading

_{Reviews (1): Last reviewed commit: "fix(bedrock): skip synthetic tool inject..." | Re-trigger Greptile}

[greptile-apps]

Duplicate import

_get_openapi_url is imported twice in the same block (lines 496 and 499). Python silently accepts this, but the duplicate is dead code and should be removed.

Suggested change

	_get_openapi_url,
	_get_projected_spend_over_limit,

[greptile-apps]

Inline import inside module-level function

Per the project style guide (CLAUDE.md), imports belong at the top of the file. The commit history notes "refactor: module level imports for ensure_litellm_metadata" was an explicit intent, but the import was left inside the function body. Move it to the top of the file unless a circular-import comment explains why it must be deferred.

Suggested change

	from litellm.llms.base_llm.guardrail_translation.base_translation import (
	BaseTranslation,
	)
	user_metadata = BaseTranslation.transform_user_api_key_dict_to_metadata(

Context Used: CLAUDE.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[greptile-apps]

Missing blank line before top-level function

PEP 8 and the surrounding code style require two blank lines between top-level definitions. Only one blank line separates _get_docs_url and _get_openapi_url.

Suggested change

	def _get_openapi_url() -> Optional[str]:
	def _get_openapi_url() -> Optional[str]:

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!