v3.12.0rc0

MLflow 3.12.0rc0 brings powerful new capabilities for agent developers, starting with automatic tracing for more AI coding assistants. First-class tracing is now available for Claude Code, Codex, Qwen Code, and Gemini CLI, distributed as standalone TypeScript-based plugins installable as CLI binaries. Users can drop one into their CLI of choice and capture every prompt, tool call, and turn out-of-the-box. Additionally, an OpenClaw tracing plugin captures every prompt, tool call, and response across sessions, allowing inspection of agent behavior directly in the MLflow UI.

The release also introduces AI Gateway guardrails with a new Guardrail base class and JudgeGuardrail implementation. Users can configure pre-LLM and post-LLM guardrails from the Gateway endpoint editor with full DB persistence and REST API support. Multimodal trace attachments enable viewing of images, audio, and files directly in the UI with click-to-expand modals and audio playback. A new mlflow.diffusers flavor provides first-class support for diffusion models, allowing users to save and serve LoRA adapters for image-generation pipelines with the same ergonomics as PyTorch or Transformers.