Issues

• [Bug]: Credential create/update/delete can be undone in memory by a concurrent reload (credential endpoints don't take MODEL_RECONCILE_LOCK)

  BerriAI/litellm#50273 · Filter by author c3-jayatinaik opened on Nov 5, 2026

• [Feature]: Headroom guardrail should use Headroom session mode so Claude Code traffic is compressible without busting prompt cache

  claude codellm translation
  BerriAI/litellm#50270 · Filter by author ppsplus-bradh opened on Nov 5, 2026

• [Bug]: Guardrail write-back on /v1/messages drops tool_result.is_error and system output_config (seen with Headroom guardrail)

  claude codellm translation
  BerriAI/litellm#50269 · Filter by author ppsplus-bradh opened on Nov 5, 2026

• [Feature]: Inbuilt credential masking guardrails for OpenAI, Anthropic, and Gemini API keys

  enhancementllm translation
  BerriAI/litellm#50266 · Filter by author yogeshwaran10 opened on Nov 5, 2026

• [Bug]: COHERE_API_BASE is ignored for cohere embeddings (rerank honors it)

  llm translation
  BerriAI/litellm#50248 · Filter by author zhengqiuyang opened on Nov 5, 2026
  1

• [Bug]: Chat→Responses bridge puts system prompts longer than 1,048,576 characters into instructions, which OpenAI rejects

  bugllm translation
  BerriAI/litellm#50243 · Filter by author mpetaccia opened on Nov 5, 2026
  1

• [Bug]: Bedrock adapter turns encrypted agent messages into text and breaks task delivery

  llm translation
  BerriAI/litellm#50224 · Filter by author strantalis opened on Nov 5, 2026
  2

• [Bug]: vertex_ai text-to-speech ignores vertex_location when building the endpoint (always global unless api_base is the full URL)

  llm translation
  BerriAI/litellm#50219 · Filter by author julienparenti opened on Nov 5, 2026
  3

• [Bug]: ui_access_mode: "admin_only" inserts uninvited SSO users into LiteLLM_UserTable and mints session keys before returning 401

  BerriAI/litellm#50216 · Filter by author JGrubb opened on Nov 5, 2026

• OpenAI workload identity tokens (#39966) can't be used for batches: OpenAI runs them with the expired creating token

  llm translation
  BerriAI/litellm#50185 · Filter by author ThisWaySir opened on Nov 5, 2026

• [Feature]: Configurable TTFT timeout across streaming endpoints

  llm translation
  BerriAI/litellm#50176 · Filter by author Achooo opened on Nov 5, 2026

• [Bug]: /v1/images/edits on vertex_ai Gemini with no image returns 500 (ValueError not mapped to 400)

  llm translation
  BerriAI/litellm#50173 · Filter by author mihidumh opened on Nov 5, 2026
  2

• [Bug]: extra_body.input overrides the Responses input that guardrails scan

  llm translation
  BerriAI/litellm#50167 · Filter by author mac146 opened on Nov 5, 2026
  1

• BitBucket prompt store ignores the configured base_url and always calls api.bitbucket.org

  BerriAI/litellm#50158 · Filter by author soarkit opened on Nov 5, 2026

• [Bug]: OpenRouter streaming emits an empty post-finish choice without usage before the final usage chunk (v1.108.0)

  bugllm translation
  BerriAI/litellm#50156 · Filter by author vovkannm opened on Nov 5, 2026
  3

• [Bug]: Requests blocked by a pre_call guardrail are stored in spend logs with the text that got them blocked

  bugllm translation
  BerriAI/litellm#50127 · Filter by author matthewbuglass-andgosystems opened on Nov 4, 2026

• [Bug]: max_budget admits calls after spend has crossed it (11 calls on a budget that fits 9)

  BerriAI/litellm#50125 · Filter by author NAMAN-VERMA-02 opened on Nov 4, 2026
  1

• [Bug]: Key budget is not charged when the provider's response has missing or malformed usage

  llm translation
  BerriAI/litellm#50124 · Filter by author NAMAN-VERMA-02 opened on Nov 4, 2026

• [Feature]: add service_tier label to litellm_proxy_failed_requests_metric

  llm translation
  BerriAI/litellm#50113 · Filter by author jimmychen-p72 opened on Nov 4, 2026

• [Bug]: Claude on Bedrock behind the OpenAI Agents SDK: deferred tool not found after tool search, search record lost on later turns, compaction param dropped on /v1/messages

  llm translation
  BerriAI/litellm#50104 · Filter by author homenshum-cheiron opened on Nov 4, 2026
  1

• Complexity Router: Claude Code Subagent Overwrites Parent's user_turn Tier Pin (Not Covered by #48701)

  claude codellm translation
  BerriAI/litellm#50099 · Filter by author ppsplus-bradh opened on Nov 4, 2026
  1

• [Docs/Bug]: health check routing does not exclude a dead deployment until the first background health check finishes, so the first requests after startup pay the full timeout

  llm translation
  BerriAI/litellm#50084 · Filter by author Atsu1979 opened on Nov 4, 2026
  2

• [Bug]: /v1/messages to Vertex Claude sends messages before system/tools, and the prompt cache misses on tool-loop rounds

  bugllm translation
  BerriAI/litellm#50070 · Filter by author brandonnovick opened on Nov 4, 2026
  2

• [Bug]: v1.108.0 MCP tools/list blocks 18-50 s on legacy subscription listener after notifications/initialized

  BerriAI/litellm#50065 · Filter by author gruuna-bannert opened on Nov 4, 2026
  3

• Can the provider logo in Model Management be customized for models routed through an OpenAI/Anthropic-compatible gateway?

  llm translation
  BerriAI/litellm#50056 · Filter by author jweisman11 opened on Nov 4, 2026