●Change8

v3.13.0rc0

Breaking Changes
📦 mlflowView on GitHub →
⚠ 2 breaking✨ 9 features🔧 6 symbols

Summary

MLflow 3.13.0rc0 introduces a major overhaul of Role-Based Access Control (RBAC) with a new Admin UI, integrates several new coding agents via plugins in the AI Gateway, and provides a production-ready Helm chart for Kubernetes deployment.

⚠️ Breaking Changes

  • Legacy per-resource permission tables have been collapsed into the unified `role_permissions` structure as part of RBAC Phase 2. Existing permission configurations relying on the old table structure may need migration.
  • The legacy Python autolog hook for `mlflow autolog claude` is replaced by the new official plugin mechanism for coding agents. Users relying on the old hook must switch to the new plugin setup.

Migration Steps

  1. Review and update existing RBAC permission configurations to use the new `role_permissions` structure instead of legacy per-resource permission tables.
  2. Migrate from the legacy `mlflow autolog claude` Python hook to the new official plugin mechanism for Claude coding agent integration.

✨ New Features

  • Major overhaul of Role-Based Access Control (RBAC) including collapsing legacy permission tables, unifying per-user permission APIs under `/mlflow/users/permissions/*`, promoting `prompt` to a first-class RBAC resource type, and seeding default roles on workspace creation.
  • New 4-page Admin UI introduced for workspace managers, covering account widget, `/account` page, Platform Admin pages, and backend auth endpoints.
  • AI Gateway now supports Claude Code, OpenClaw, Ollama, and OpenAI Codex as first-class assistant providers via new plugins.
  • A Claude Code TypeScript plugin with a setup wizard and `settings.local.json` support has been added.
  • A coding-agent endpoint creation flow is now available directly in the AI Gateway UI.
  • First-class Helm chart provided for deploying MLflow to Kubernetes clusters, including production-ready configuration, ingress, and persistence.
  • New API `mlflow.genai.test_agent` introduced for stress-testing GenAI agents by generating adversarial inputs and replaying them.
  • Tracing now supports the OpenTelemetry Link entity via `LiveSpan.add_link()`, allowing connection of causally related spans across traces.
  • SQL tracking store now supports reader/writer instance routing for database replicas to enable horizontal scaling for read-heavy deployments.

Affected Symbols

/mlflow/users/permissions/*`role_permissions``prompt` (as RBAC resource_type)`mlflow autolog claude``LiveSpan.add_link()`AI Gateway UI (for endpoint creation)