ml-party

An agent-native platform for ML experiment tracking and lineage/knowledge.

A durable, agent-legible knowledge substrate for ML work — a lab notebook that agents write (over MCP) and query, and that humans read live (CLI + web UI) — that also happens to be a full local tracker: metrics, artifacts, checkpoints, reproducibility, all stored locally.

Linux/macOS (Windows via WSL), Python ≥ 3.11. The web UI ships prebuilt in the wheel:

pip install mlparty

mlp init --root .mlparty                    # create a store (+ MCP registration)
python -m mlparty.demo &                    # a real run: contract + live metrics
mlp ui                                      # → http://127.0.0.1:7327

To develop against a checkout instead, see From source (adds Node ≥ 20 for the UI build).

From there: agents pick up the MCP server automatically and it teaches them the workflow (MCP setup & tools); training scripts attach as the second writer (Tracking runs); humans watch live in the web UI (The web UI); registered actions let agents drive jobs, audited (Run control (actions)).

User guide

• Tracking runs
  • The two writers
  • The write contract
  • What gets captured automatically
  • Heartbeats
  • RunHandle API
• The web UI
  • The metrics dashboard
  • The artifacts tab
  • Live streaming
• Boards
  • Creating one
  • The board contract
  • Boards under authentication
  • Security model
  • Boards and the knowledge graph
• Run control (actions)
  • Registering a template
  • Invoking
  • Restart is never a mutation
  • Security model
• MCP setup & tools
  • Zero-terminal setup
  • Tools (18)
  • Code capture is asked for, never assumed
• Remote tracking (spool-and-flush)
  • Server (machine S)
  • Client (machine C or L)
  • What ships, and why it’s safe
  • Caveats
• Deployment
  • Auth: users, roles, tokens
  • Decision tree
  • Docker
  • Compose
  • Backup & restore
  • Security model
  • Who needs access to what

Reference

• API reference
• Design
• ml-party — Design
• Changelog