Home

OpenRunner

Open-source ML experiment tracking platform — a self-hosted alternative to Weights & Biases.

Live demo: https://openrun.gladia.io | PyPI: pip install openrunner-sdk

Quick Links

• Installation
• Quick Start
• SDK Reference
• Session Sync
• Blog Builder
• Paper Builder
• Auto Code Capture
• MCP Integration
• Integrations
• Self-Hosting
• REST API
• Configuration
• Architecture

Features

• Full W&B API compatibility (drop-in replacement)
• 30 framework integrations (PyTorch, HuggingFace, LangChain, etc.)
• Real-time experiment comparison with charts
• Artifact versioning with dedup, TTL, lineage, and model registry
• Blog Builder — AI-generated posts with Vega-Lite charts, inline comments (select + c), chat editor, revision history, full project context
• Paper Builder — LaTeX editor with rendered preview, inline annotations, chat editor, multi-model debate, import .tex from artifacts
• 13 MCP tools for Claude Code (search, upload artifacts, edit papers, watch runs)
• 13 Claude Code slash commands (push/pull papers, sync sessions, plot metrics)
• LLM tracing with OpenTelemetry ingestion
• Evaluation framework with LLM-as-judge scorers + WER scorer
• Guardrails, annotation queues, feedback system
• Case-insensitive org/project routing
• Light/dark/system theme
• Self-hosted with Docker Compose or Kubernetes (Helm chart included)
• TimescaleDB for time-series metric storage