●Change8

litellm_langfuse-dev-v1.81.13

📦 litellmView on GitHub →
✨ 7 features🐛 44 fixes🔧 8 symbols

Summary

This release introduces a new Prompt Management API and significant UI enhancements, including Access Group selection and Usage filtering. It also contains extensive bug fixes focused heavily on improving test stability and isolation across various integrations.

Migration Steps

  1. If using reasoning functionality, ensure the reasoning parameter is set to None instead of Reasoning() where appropriate.
  2. If relying on policy_resolve_endpoints, ensure your environment does not strictly require FastAPI to be present for imports.

✨ New Features

  • Introduced a new Prompt Management API for interacting with Prompt Management integrations.
  • Added Access Group Selector to Create and Edit Flow in the UI (Keys/Teams section).
  • Added filtering by User in the Usage UI.
  • Implemented Guardrail tracing UI showing policy, detection method, and match details.
  • Added configurable fallback mechanism for generic guardrail endpoint connection failures.
  • Added EU AI Act Article 5 policy template for prohibited practices detection.
  • Added GDPR Art. 32 EU PII Protection Policy Template.

🐛 Bug Fixes

  • Preserved key_alias and team_id metadata in /user/daily/activity/aggregated after key deletion or regeneration.
  • Fixed an issue where the reasoning parameter was incorrectly using Reasoning() instead of None.
  • Removed unused Reasoning import from transformation.py.
  • Fixed flaky tests related to Langfuse isolation.
  • Resolved merge conflict and fixed Bedrock thinking test flakiness.
  • Fixed flaky tests caused by module reloading and missing mocks.
  • Fixed SSO test flakiness by correctly mocking premium_user.
  • Made policy_resolve_endpoints importable without FastAPI.
  • Improved Langfuse test isolation to prevent flaky failures.
  • Added mock isolation for test_video_content_handler_uses_get_for_openai.
  • Added cleanup for disable_aiohttp_transport in test_extra_body_with_fallback.
  • Restored Langfuse client counter in test cleanup.
  • Removed redundant cache flush from test_openai_env_base.
  • Added environment cleanup for Vertex AI rerank tests.
  • Updated reasoning_effort test to expect dict format.
  • Added environment cleanup for Vertex AI GPT-OSS tests.
  • Added environment cleanup for Vertex AI Qwen tests.
  • Used async side_effect for client.post mock in watsonx test.
  • Mocked vertexai module in GPT-OSS tests to prevent authentication issues.
  • Updated test_other_constraints_preserved for new schema filtering.
  • Added fakeredis dependency for pod lock manager tests.
  • Cleared tokenizer LRU cache for test isolation.
  • Mocked environment variables for callback validation test.
  • Fixed unsupported parameters for chatgpt codex.
  • Fixed converse anthropic usage object according to v1/messages specs.
  • Applied miscellaneous bug fixes for Managed batches.
  • Clamped thinking.budget_tokens to a minimum of 1024 for Bedrock.
  • Fixed virtual key grace period retrieval from env/UI.
  • Fixed SSO PKCE support failure in multi-pod Kubernetes deployments.
  • Added pytest-postgresql dependency for db schema migration tests.
  • Replaced caplog with custom handler for parallel execution in tests.
  • Corrected async mock for video generation logging test.
  • Added cleanup fixture and no_parallel mark for MCP tests.
  • Added a fallback to handle generic guardrail endpoint connection failures.
  • Excluded tool params for models without function calling support.
  • Preserved metadata for custom callbacks on codex/responses path.
  • Handled missing DATABASE_URL in append_query_params for the proxy.
  • Reverted StreamableHTTPSessionManager to stateless mode in MCP.
  • Prevented double-counting of litellm_proxy_total_requests_metric.
  • Returned finish_reason='tool_calls' when response.completed contains function_call items in responses-api.
  • Fixed OCI Grok output pricing.
  • Fixed master key rotation Prisma validation errors in the proxy.
  • Made vector stores migration idempotent.
  • Fixed UI issues related to Content Filters (help edit/view categories, 1-click add categories, go to next page).

Affected Symbols

policy_resolve_endpointsReasoningkey_aliasteam_idStreamableHTTPSessionManagerlitellm_proxy_total_requests_metricresponse.completedfinish_reason