MLflow 3.7.0 includes several major features and improvements for GenAI Observability, Evaluation, and Prompt Management.

Major Features​

• 📝 Experiment Prompts UI: New prompts functionality in the experiment UI allows you to manage and search prompts directly within experiments, with support for filter strings and prompt version search in traces. (#19156, #18919, #18906, @TomeHirata)
• 💬 Multi-turn Evaluation Support: Enhanced mlflow.genai.evaluate now supports multi-turn conversations, enabling comprehensive assessment of conversational AI applications with DataFrame and list inputs. (#18971, @AveshCSingh)
• ⚖️ Trace Comparison: New side-by-side comparison view in the Traces UI allows you to analyze and debug LLM application behavior across different runs, making it easier to identify regressions and improvements. (#17138, @joelrobin18)
• 🌐 Gemini TypeScript SDK: Auto-tracing support for Google's Gemini in TypeScript, expanding MLflow's observability capabilities for JavaScript/TypeScript AI applications. (#18207, @joelrobin18)
• 🎯 Structured Outputs in Judges: The make_judge API now supports structured outputs, enabling more precise and programmatically consumable evaluation results. (#18529, @TomeHirata)
• 🔗 VoltAgent Tracing: Added auto-tracing support for VoltAgent, extending MLflow's observability to this AI agent framework. (#19041, @joelrobin18)

Breaking Changes​

• [Tracking] SQLite is now the default backend for the MLflow Tracking server. (#18497, @harupy)
• [Models] Remove deprecated diviner flavor (#18808, @copilot-swe-agent)
• [Models] Remove deprecated promptflow flavor (#18805, @copilot-swe-agent)

Features​

• [Tracking] Create parent directories for SQLite database files (#19205, @harupy)
• [Prompts] Link Prompts and Experiments when prompts are loaded/registered (#18883, @TomeHirata)
• [Tracking] Include environment variable fallback for SGC run resumption (#19143, @artjen)
• [Tracking] Add support for SGC run resumption from Databricks Jobs (#19015, @artjen)
• [Evaluation] Add --builtin/-b flag to mlflow scorers list command (#19095, @alkispoly-db)
• [Tracing] Pydantic AI Chat UI support (#18777, @joelrobin18)
• [Tracking] Add auth support for scorers (#18699, @BenWilson2)
• [Evaluation] Remove experimental flags from scorers (#18122, @BenWilson2)
• [Evaluation] Add description field to all built-in scorers (#18547, @alkispoly-db)

Bug Fixes​

• [Tracing] Handle traces with third-party generic root span (#19217, @B-Step62)
• [Tracing] Fix OTLP endpoint path handling per OpenTelemetry spec (#19154, @harupy)
• [Tracing] Add gzip/deflate Content-Encoding support to OTLP traces endpoint (#19024, @Miaoxiang-philips)
• [Tracing] Add missing _delete_trace_tag_v3 API (#18813, @Tian-Sky-Lan)
• [Tracing] Fix bug in chat sessions view where new sessions created after UI launch are not visible due to incorrect timestamp filtering (#18928, @dbczumar)
• [Tracing] Fix OTLP proto conversion for empty list/dict (#18958, @B-Step62)
• [Tracing] Agno V2 fixes (#18345, @joelrobin18)
• [Tracing] Fix /v1/traces endpoint to return protobuf instead of JSON (#18929, @copilot-swe-agent)
• [Tracing] Pin click!=8.3.0 in MCP extra to fix MCP server failure (#18748, @copilot-swe-agent)
• [Tracing] Fix MCP server uv installation command for external users (#18745, @copilot-swe-agent)
• [Evaluation] Fix trace-based scorer evaluation by using agentic judge adapter (#19123, @alkispoly-db)
• [Evaluation] Fix managed scorer registration failure (#19146, @xsh310)
• [Evaluation] Fix InstructionsJudge using scorer description as assessment value (#19121, @alkispoly-db)
• [Evaluation] Add validation to correctness judge expectation fields (#19026, @smoorjani)
• [Evaluation] Fix model URI underscore handling (#18849, @RohanRouth)
• [Evaluation] Fix evaluate_traces MCP tool error: use result_df instead of tables (#18825, @alkispoly-db)
• [Evaluation] Fix Bedrock Anthropic adapter by adding required anthropic_version field (#17744, @harupy)
• [Evaluation] Fix migration for pre-existing auth tables (#18793, @BenWilson2)
• [Tracking] Fix tracking URI propagation (#18023, @shaperilio)
• [Tracking] Fix SqlLoggedModelMetric association with experiment_id (#18382, @mcompen)
• [Tracking] Add Flask routes to auth validators (#18486, @BenWilson2)
• [Tracking] Add missing proto handler for Experiment association handling for datasets (#18769, @BenWilson2)
• [UI] Show full dataset record content and add search bar in evaluation datasets UI (#19000, @dbczumar)
• [UI] Request TraceInfo and Trace Assessments from a relative API path (#19032, @kbolashev)
• [UI] Define LoggedModelOutput.to_dictionary() so LoggedModelOutput and runs containing them can be JSON serialized (#19017, @nicklamiller)
• [UI] Fix router issue in TracesUI page (#19044, @joelrobin18)
• [Build] Fix mlflow gc to remove model artifacts (#17282, @joelrobin18)
• [Build] Fix Click 8.3.0 Sentinel.UNSET handling in MCP server (#18858, @harupy)
• [Build] Add bucket-ownership checks for Amazon S3 (#18542, @kingroryg)
• [Docs] Fix Python indentation in custom trace quickstart example (#19185, @copilot-swe-agent)
• [Docs] Fix property blocks rendering horizontally in API documentation (#19125, @copilot-swe-agent)
• [Docs] Fix CLI link missing api_reference prefix in documentation sidebars (#18893, @copilot-swe-agent)
• [Docs] Fix notebook download URLs to use versioned paths (#18806, @harupy)
• [Docs] Fix documentation redirects for removed getting-started pages (#18789, @copilot-swe-agent)
• [Models] Fix shared cluster Py4j statefulness issue (#19139, @BenWilson2)
• [Models] Prevent symlink path traversal in local artifact store (#18964, @BenWilson2)

Documentation Updates​

• [Docs] Add LangGraph optimization guide (#19180, @TomeHirata)
• [Docs] Add documentation for milestone 1 of multi-turn evaluation support (#19033, @smoorjani)
• [Docs] Update transformers and sentence transformers docs (#18925, @BenWilson2)
• [Docs] Clean up Classic Eval docs (#19013, @BenWilson2)
• [Docs] Improve documentation for prompt_template (#19105, @ingo-stallknecht)
• [Docs] Fix typos in ML documentation main page (#19048, @copilot-swe-agent)
• [Docs] Convert documentation GIF animations to MP4 videos (#18946, @harupy)
• [Docs] Improve readability by adjusting sidebar layout and style (#18937, @kevin-lyn)
• [Docs] Clean up scikit-learn docs (#18794, @BenWilson2)
• [Docs] Clean up XGBoost docs (#18790, @BenWilson2)
• [Docs] Clean up TensorFlow docs (#18850, @BenWilson2)
• [Docs] Use the correct OTLP HTTP exporter in OTel collector YAML (#18930, @Miaoxiang-philips)
• [Docs] Clean up SpaCy and Keras docs (#18895, @BenWilson2)
• [Docs] Fix contents in tracing doc pages (#18750, @B-Step62)
• [Docs] Improve file store deprecation warning messages (#18900, @harupy)
• [Docs] Clean up the MLflow 3 docs content (#18871, @BenWilson2)
• [Docs] Add multi-turn judge creation with make_judge API and direct judge invocation (#18897, @xsh310)
• [Docs] Clean up PyTorch docs (#18816, @BenWilson2)
• [Docs] Clean up Prophet docs (#18814, @BenWilson2)
• [Docs] Clean up SparkML docs (#18811, @BenWilson2)
• [Docs] Clean up the traditional ML landing page (#18799, @BenWilson2)
• [Docs] Clean up the Deep Learning landing page (#18820, @BenWilson2)
• [Docs] Clean up evaluation datasets docs (#18766, @BenWilson2)
• [Docs] Fix OpenTelemetry documentation (#18810, @joelrobin18)
• [Docs] Clarify mlflow gc command behavior for pinned runs and registered models (#18704, @copilot-swe-agent)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.