fix(litellm_proxy): add "fake-api-key" fallback when LITELLM_PROXY_API_KEY is not set

[ritsuki1227]

Changes

Add or "fake-api-key" fallback to the litellm_proxy provider, matching the existing pattern in hosted_vllm, lm_studio, llamafile, and datarobot.

This enables keyless authentication scenarios (service mesh with mTLS, VPC-internal, local development) where the LiteLLM proxy runs with master_key=None.

Why "fake-api-key" and not ""?

The fallback value must be truthy. In main.py:2366-2371, there is an or-chain:
api_key = api_key or litellm.api_key or litellm.openai_key or get_secret("OPENAI_API_KEY")
An empty string "" is falsy and would fall through to OPENAI_API_KEY, potentially leaking unrelated credentials to the proxy. "fake-api-key" is truthy and stops the chain.

Relevant issues

Fixes #20925

Pre-Submission checklist

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

• I have Added testing in the tests/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

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:

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	Feb 22, 2026 7:14am

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Overview

Greptile Summary

Added "fake-api-key" fallback across all four litellm_proxy sub-providers (chat, image_edit, image_generation, responses) so the proxy can be used without setting LITELLM_PROXY_API_KEY. This matches the established pattern in hosted_vllm, lm_studio, llamafile, and datarobot providers. The truthy "fake-api-key" value prevents the or-chain in main.py from falling through to OPENAI_API_KEY, avoiding credential leakage to the proxy.

• The responses sub-provider previously had no validate_environment override and was inheriting OpenAIResponsesAPIConfig's default, which would resolve to OPENAI_API_KEY — a bug when the proxy doesn't need authentication. A new override was added with the correct provider-specific fallback chain.
• Whitespace/formatting cleanups in image_generation/transformation.py and responses/transformation.py.
• Comprehensive parametrized tests added for all four sub-providers, covering user-provided keys, env var keys, and the "fake-api-key" fallback. Two additional E2E mock tests verify that OPENAI_API_KEY is never leaked to the proxy.

Confidence Score: 5/5

• This PR is safe to merge — it applies a well-established fallback pattern consistently across the litellm_proxy provider with thorough test coverage.
• The change is minimal and follows an identical pattern already used in 5+ other providers. The method signatures match parent classes correctly. Tests are comprehensive, mock-only, and cover all edge cases including credential isolation. No breaking changes or performance concerns.
• No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/litellm_proxy/chat/transformation.py	Added or "fake-api-key" fallback to _get_openai_compatible_provider_info and get_api_key methods, plus a docstring correction. Follows the exact same pattern as hosted_vllm, lm_studio, and llamafile providers.
litellm/llms/litellm_proxy/image_edit/transformation.py	Added or "fake-api-key" fallback to validate_environment. Straightforward change consistent with other litellm_proxy transformation files.
litellm/llms/litellm_proxy/image_generation/transformation.py	Added or "fake-api-key" fallback to validate_environment and a minor formatting fix (blank line after docstring). Consistent with other files.
litellm/llms/litellm_proxy/responses/transformation.py	Added new validate_environment override that uses LITELLM_PROXY_API_KEY with "fake-api-key" fallback instead of inheriting OpenAI's default which would fall through to OPENAI_API_KEY. Method signature correctly matches parent class OpenAIResponsesAPIConfig.validate_environment.
tests/test_litellm/llms/litellm_proxy/chat/test_litellm_proxy_chat_transformation.py	Added well-structured parametrized tests for _get_openai_compatible_provider_info and get_api_key, plus two E2E mock tests verifying the fake-api-key fallback and OPENAI_API_KEY isolation. All tests use mocks only.
tests/test_litellm/llms/litellm_proxy/image_edit/test_litellm_proxy_image_edit_transformation.py	New test file with parametrized tests covering all API key fallback scenarios for image edit. Mock-only, no network calls.
tests/test_litellm/llms/litellm_proxy/image_generation/test_litellm_proxy_image_generation_transformation.py	New test file with parametrized tests covering all API key fallback scenarios for image generation. Mock-only, no network calls.
tests/test_litellm/llms/litellm_proxy/responses/test_litellm_proxy_responses_transformation.py	New test file with parametrized tests plus an explicit test that OPENAI_API_KEY is not leaked to the proxy. Mock-only, no network calls.

[greptile-apps]

_{8 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[ritsuki1227]

@krrishdholakia This PR is ready for review.

CI note: The only failing checks are repo-wide issues unrelated to this PR:

• lint: litellm_pre_call_utils.py (PLR0915) and pipeline_executor.py (F841) — not touched by this PR
• test matrix: poetry.lock / pyproject.toml mismatch — affects all open PRs
• unit-test and validate-model-prices-json both pass

Also note that #20947 appears to be a duplicate of this PR, covering only the chat sub-provider. This PR was opened earlier and covers all 4 litellm_proxy sub-providers (chat, image_edit, image_generation, responses) with tests for each.

[ritsuki1227]

@ishaan-jaff @krrishdholakia Friendly bump — rebased onto latest main, CI is green now.

This is a small fix (1 commit, 8 files) adding the "fake-api-key" fallback to the litellm_proxy provider, matching the existing pattern in hosted_vllm, lm_studio, llamafile, and datarobot. Fixes #20925.

[github-actions]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.