Shipped · Observability

What mlflow/mlflow shipped

The public repository of Mlflow · github.com/mlflow/mlflow

Written by FoxPlug from the repository's own releases; not affiliated with Mlflow. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/mlflow/mlflow; Mlflow did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.

Get a weekly update like this for your product, free

Or Use it as a GitHub Action

Follow mlflow's weekly shipped digest

Week of October 5, 2026

What shipped

Released MLflow 3.17.0 with Jev Decisions across evaluation, gateway and tracing, plus fine-grained permissions for shared resources. Release
TracedOpenAI() in TypeScript now recognizes Decisions.create() and records native questions and answers in an LLM span. Pull request #26576
Fixed nested object properties being stripped from function tool schemas so the gateway validates tool parameters correctly. Pull request #26646
Precision_at_k, recall_at_k and ndcg_at_k now accept all NumPy integer dtypes for document IDs. Pull request #26643
Fixed tensor signature enforcement failing for DataFrames whose row labels are not 0, such as a filtered RangeIndex. Pull request #26462
Added mlflow.genai.databricks.review_queues, an experimental client for the Databricks review queue REST API covering questions, queues, members and items. Pull request #26414

Changelog entry

Released MLflow 3.17.0 with Jev Decisions across evaluation, gateway and tracing, and fine-grained permissions for shared resources. Release
Python autolog now traces synchronous and asynchronous decisions.create() calls. Pull request #26575
Fixed nested object properties being stripped from function tool schemas. Pull request #26646
Precision_at_k, recall_at_k and ndcg_at_k accept all NumPy integer dtypes for document IDs. Pull request #26643
Fixed tensor schema enforcement for DataFrames with non-default indexes. Pull request #26462
Log traceback when GenAI semconv message conversion fails in translate_span_to_genai. Pull request #26632
Added mlflow.genai.databricks.review_queues for managing Databricks review queues. Pull request #26414

Week of September 28, 2026

What shipped

Added TypeSafe AI System One tracing integration to the MLflow TypeScript SDK with the @mlflow/typesafe package and tracedTypeSafe() client wrapper. Pull request #26342
Added MLflow Tracing autologging support for TypeSafe AI System One through mlflow.typesafe.autolog(). Pull request #26129
Added documentation page for Custom Trace Views, a headline feature of the 3.16.0 release. Pull request #26356
Added Skill Registry discovery catalog UI with lazy routes for the Skills entry. Pull request #26274
Added TraceImage component to custom trace views for rendering image attachments from traces. Pull request #26297
Fixed GenAI semconv spans to keep tool and agent names in execute_tool and invoke_agent span names. Pull request #26394
Fixed mlflow.artifacts.load_text and load_dict to read artifacts as UTF-8 instead of locale encoding. Pull request #26265
Improved server responsiveness by serving GET /api/3.0/mlflow/server-info on dedicated FastAPI routes instead of shared WSGI thread pool. Pull request #26267

Why it matters

This week adds significant TypeScript SDK capabilities with TypeSafe AI integration and autologging, improves documentation for custom tracing features, and fixes multiple streaming issues in tool call handling across providers. Several artifact and endpoint fixes address data handling and UTF-8 encoding problems.

Changelog entry

Added @mlflow/typesafe tracing integration package for TypeSafe AI System One to MLflow TypeScript SDK Pull request #26342
Added TypeSafe AI System One autologging support via mlflow.typesafe.autolog() Pull request #26129
Added Custom Trace Views documentation page Pull request #26356
Added Skill Registry discovery catalog UI Pull request #26274
Added TraceImage component for rendering image attachments in custom trace views Pull request #26297
Fixed GenAI semconv spans to preserve tool and agent names in span names Pull request #26394
Fixed load_text and load_dict to read artifacts as UTF-8 Pull request #26265
Fixed GeminiAdapter to return 422 for unmatched tool_call_id Pull request #26253
Improved server responsiveness by serving /api/3.0/mlflow/server-info on dedicated FastAPI routes Pull request #26267

Week of September 21, 2026

What shipped

MLflow 2.11.5 released with opt-in support for routing Unity Catalog model registry artifact uploads and downloads through the Databricks SDK Files API. Release
SQL daily rollups and optimizations for trace analytics implemented as opt-in feature, with authoritative SQL columns and migration/backfill support. Pull request #25991
Initial phase of RFC-0002 merged to enable job executor plugins, allowing extensible job execution strategies. Pull request #26119
Skill Registry now supports secure skill content upload and artifact lifecycle management with audit metadata tracking. Pull request #25958
Added end-to-end support for uc_model_service model-serving resource type in Unity Catalog model-version registration. Pull request #26140
Session metadata now propagates to TypeScript UC root spans, fixing session filtering for Unity Catalog-backed traces. Pull request #26209
DataFrame indexes are preserved during pyfunc schema enforcement for Array, Object, and Map columns. Pull request #26181
Fixed metric filter in search_logged_models that was shrinking result pages by applying subquery joins incorrectly. Pull request #26057
Spaces around --trace-ids entries in mlflow traces delete are now stripped to handle comma-separated lists correctly. Pull request #26204

Why it matters

This week brings infrastructure improvements for trace analytics and job execution extensibility, alongside fixes for common issues in model registration, tracing across multiple LLM providers, and data handling in pandas DataFrames. The SQL optimizations and plugin architecture lay groundwork for scaling production deployments.

Changelog entry

Added opt-in support for routing Unity Catalog model registry artifact uploads and downloads through the Databricks SDK Files API Release
Added opt-in SQL daily rollups and optimizations for trace analytics including authoritative SQL columns and migration/backfill support Pull request #25991
Merged initial phase of RFC-0002 to support job executor plugins for extensible job execution Pull request #26119
Added secure skill content upload and artifact lifecycle management to Skill Registry with audit metadata tracking Pull request #25958
Added end-to-end support for uc_model_service resource type in Unity Catalog model-version registration Pull request #26140
Session metadata now propagates to TypeScript UC root spans for proper session filtering in Unity Catalog traces Pull request #26209
DataFrame indexes are preserved during pyfunc schema enforcement for Array, Object, and Map columns Pull request #26181
Fixed metric filter in search_logged_models that was shrinking pages through incorrect subquery joins Pull request #26057
Fixed mlflow traces delete --trace-ids to properly strip spaces around comma-separated entries Pull request #26204

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

MLflow 2.11.5: opt-in UC model registry artifact routing via Databricks SDK Files API, SQL trace analytics optimizations, job executor plugins, and Skill Registry artifact lifecycle support.

Week of September 14, 2026

What shipped

MLflow 3.16.1 patch release with bug fixes, documentation updates, and security fixes including removal of the default basic-auth admin password. Release
Fixed security vulnerability that prevented the tracking server from accepting arbitrary artifact location URIs outside the default artifact root. Pull request #25891
Added validation to reject conflicting protobuf field spellings in request bodies to prevent bypassing security checks. Pull request #25869
Fixed third-party scorer module matching to use exact module names instead of package prefix matching, closing a potential security gap. Pull request #25906
Implemented SQL trace analytics daily rollups maintenance with scheduled periodic task execution. Pull request #25489
Backported Unity Catalog model registry Databricks Files API artifact repository support to version 2.11.4. Pull request #25986
Backported Unity Catalog model registry Databricks Files API artifact repository support to version 2.13.3. Pull request #25987
Fixed Unity Catalog trace location resolution in Databricks Model Serving to work without local-store validation. Pull request #25884
Fixed pandas Series alignment in evaluation results table to use positional indexing instead of label-based indexing. Pull request #25799

Changelog entry

[Security] Refuse to connect to client-named artifact hosts outside the default artifact root (GHSA-mr9f-g8qf-4w4j). Pull request #25891
[Security] Reject conflicting protobuf field spellings in request bodies (GHSA-3g8m-hm3x-gh2r). Pull request #25869
[Security] Bind trace_ids to authorized experiment on issues/invoke and genai/evaluate/invoke. Pull request #25873
[Traces] Maintain SQL trace analytics daily rollups with scheduled periodic task execution. Pull request #25489
[Evaluation] Fix positional Series alignment in eval_results_table for consistent index handling. Pull request #25799
[Model Serving] Resolve UC trace location without local-store validation in Databricks Model Serving. Pull request #25884
[Security] Match third-party scorer modules exactly instead of by package prefix (GHSA-26p8-2jq9-3vq9). Pull request #25906

Example posts FoxPlug drafted from these changes. Not written or posted by the project.

Post for X

MLflow 3.16.1 is out with bug fixes, security updates, and improvements to gateway model discovery and UC trace handling.

Weeks with too little public activity are left out rather than filled in. Last updated 2026-10-12.

Is this your repo? Ask us to remove this page.