Add dedicated xai_key and fallback logic for xAI API key

[PeterDaveHello]

Relevant issues

Replaces #18660 because GitHub would not allow reopening it after the branch
was force-pushed/recreated.

Addresses #18660 (comment).

Linear ticket

N/A

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
  • Ran: .venv/bin/pytest tests/test_litellm/llms/xai/test_xai_key_fallback.py -q
  • Ran: .venv/bin/pytest tests/test_litellm/llms/xai/test_xai_chat_transformation.py tests/test_litellm/llms/xai/responses/test_xai_responses_transformation.py -q
• 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

.venv/bin/pytest tests/test_litellm/llms/xai/test_xai_key_fallback.py -q
20 passed in 0.49s

.venv/bin/pytest tests/test_litellm/llms/xai/test_xai_chat_transformation.py tests/test_litellm/llms/xai/responses/test_xai_responses_transformation.py -q
18 passed in 1.13s

Type

🐛 Bug Fix
✅ Test

Changes

• Add litellm.xai_key fallback for xAI API key resolution
• Share xAI key resolution across chat, responses, and realtime paths
• Preserve Responses API and realtime legacy litellm.api_key before XAI_API_KEY behavior
• Add focused tests for chat, responses, realtime, resolver, and model listing fallback order

[codecov]

Codecov Report

❌ Patch coverage is 89.12387% with 36 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/router.py	87.20%	11 Missing ⚠️
litellm/cost_calculator.py	74.35%	10 Missing ⚠️
litellm/proxy/proxy_server.py	35.71%	9 Missing ⚠️
litellm/integrations/opentelemetry.py	90.69%	4 Missing ⚠️
...integrations/opentelemetry_utils/gen_ai_semconv.py	98.50%	1 Missing ⚠️
litellm/proxy/route_llm_request.py	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[greptile-apps]

Greptile Summary

This PR introduces litellm.xai_key as a dedicated, provider-specific API key for xAI, and centralizes key resolution for all xAI endpoints (chat, responses, realtime, model listing) into a single XAIModelInfo.get_api_key helper.

• New fallback slot: litellm.xai_key is inserted between an explicitly passed key and the XAI_API_KEY env var across all xAI paths, while responses and realtime endpoints preserve their historical litellm.api_key precedence over the env var via the legacy_generic_before_env flag.
• No regressions for existing callers: callers who don't set xai_key see the same effective priority chain as before; the new slot is simply None by default.
• 20 focused mock tests validate every permutation of the fallback order for the resolver, chat config, responses config, realtime, and model-listing paths.

Confidence Score: 5/5

Safe to merge — the change is purely additive, backward-compatible, and well-covered by tests.

All three xAI code paths (chat, responses, realtime) correctly delegate to the centralized resolver, which is additive by design: the new xai_key slot defaults to None, leaving existing behavior unchanged for callers who don't set it. The legacy_generic_before_env paths preserve the pre-existing litellm.api_key → XAI_API_KEY order for responses and realtime. Tests are mock-only and cover all priority-chain permutations.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/xai/common_utils.py	Adds legacy_generic_before_env flag to get_api_key, centralizing xAI key resolution with two distinct priority chains; error message improved. No logic errors found.
litellm/llms/xai/chat/transformation.py	Delegates key resolution to XAIModelInfo.get_api_key; behavior preserved for existing callers, new xai_key fallback added cleanly.
litellm/llms/xai/responses/transformation.py	Migrates key resolution to the shared helper with legacy_generic_before_env=True, preserving the historical litellm.api_key → XAI_API_KEY ordering. Updated error message is more descriptive.
litellm/realtime_api/main.py	Replaces inline 3-step fallback with XAIModelInfo.get_api_key(..., legacy_generic_before_env=True), keeping the same effective order while gaining xai_key support.
tests/test_litellm/llms/xai/test_xai_key_fallback.py	20 new mock-only tests covering all fallback orderings for the resolver, chat, responses, realtime, and model-listing paths. No real network calls.

_{Reviews (2): Last reviewed commit: "Add dedicated xai_key and fallback logic..." | Re-trigger Greptile}

[oss-pr-review-agent-shin]

🤖 litellm-agent: Auto-merge skipped — the staging branch shin_agent_oss_staging_05_16_2026 has 4 commit(s) not in your branch. Merging as-is would produce a confusing diff on the staging PR.

Please rebase your branch onto shin_agent_oss_staging_05_16_2026 and push; the agent will re-review automatically.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
5 out of 6 committers have signed the CLA.

✅ Cyberfilo
✅ FabrizioCafolla
✅ escon1004
✅ PeterDaveHello
✅ tomdee
❌ krrish-berri-2
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[PeterDaveHello]

Rebased onto shin_agent_oss_staging_05_16_2026 and pushed as requested.

[PeterDaveHello]

The requested rebase looks ... strange.

[oss-pr-review-agent-shin]

🤖 litellm-agent: Auto-merge skipped — the staging branch shin_agent_oss_staging_05_20_2026 has 7 commit(s) not in your branch. Merging as-is would produce a confusing diff on the staging PR.

Please rebase your branch onto shin_agent_oss_staging_05_20_2026 and push; the agent will re-review automatically.

[PeterDaveHello]

@Sameerlite, is there anything I can do to make it right?

[Sameerlite]

Extremely sorry about this. the branch was deleted as a part of cleanup process. Can you please create a new one with litellm_internal_staging

[PeterDaveHello]

@Sameerlite, thanks, I'm willing to, but also a little bit frustrated and afraid of the same thing happening once again. Is there any chance of preventing this situation?

[Sameerlite]

litellm_internal_staging doesn't get deleted. It's the main branch. So it should not happen

[PeterDaveHello]

Hi @Sameerlite, I opened a fresh replacement PR targeting litellm_internal_staging: #28647. Thanks again for the guidance.