Arbor

Arbor is a general-purpose autonomous research agent that, given a benchmark and objective, can propose hypotheses, edit code, run real-world experiments, and retain validated data. It continuously optimizes tasks by building hypothesis trees rather than discarding failed attempts. Arbor can be used for model training, engineering optimization, data synthesis, and more. Users can operate it via a simple command-line tool or web interface without needing to configure API keys. Arbor supports multiple backend models and integrates with Codex and Claude Code.