MLflow 3.13.0rc0 adds RBAC admin UI, coding-agent plugins, and Helm charts
New release candidate brings production-grade RBAC, Claude Code integration, and Kubernetes Helm support.
MLflow 3.13.0rc0 delivers a major RBAC overhaul (Phase 2) featuring a new 4-page Admin UI, unified user permission APIs, and workspace-level USE permissions that let users create experiments and models. Legacy per-resource permission tables are consolidated into a role_permissions schema, and default roles are seeded on workspace creation. The prompt resource type is now a first-class RBAC entity, giving platform admins granular control.
On the AI observability front, coding-agent tracing arrives as plugins: Claude Code, OpenClaw, Ollama, and OpenAI Codex are wired into the AI Gateway as first-class assistant providers. A Claude Code TypeScript plugin includes a setup wizard and settings.local.json support. For deployment, MLflow now offers a production-ready Helm chart for Kubernetes—ingress, persistence, and appVersion pre-configured—so teams can spin up a tracking server with a single helm install command. Additional highlights include a genai.test_agent API for adversarial stress-testing of GenAI agents, OpenTelemetry span links via LiveSpan.add_link() for cross-trace causal connections, and database replica routing for read-heavy workloads to scale horizontally without burdening the primary.
- RBAC Phase 2 introduces a 4-page Admin UI, unified /mlflow/users/permissions/* APIs, and workspace USE permission for creating experiments and models.
- Coding-agent tracing plugins now support Claude Code, OpenAI Codex, OpenClaw, and Ollama as first-class AI Gateway providers.
- First-class Helm charts enable one-command Kubernetes deployment with production-ready ingress, persistence, and appVersion.
Why It Matters
Streamlines MLOps with enterprise-grade access control, native agent tracing, and simplified Kubernetes deployment.