Arbor

Arbor is an autonomous research agent that runs experiments, tests hypotheses, and optimizes code.

Arbor is an autonomous research framework that manages complex experimentation loops. It uses a two-agent design with a coordinator that maintains an idea tree and executors that implement code changes in isolated git worktrees. The system supports literature search via the alphaXiv API to check idea novelty before running experiments. It works with models from Anthropic, OpenAI, and other providers via LiteLLM. Users can operate it through a native command-line interface or as a skill suite inside coding assistants.

Arbor at a glance

Best for: Researcher
Runs on: CLI, Python
Pricing: Free
Released: May 2026
Category: AI research tool
Website: github.com/RUC-NLPIR/Arbor

Key features

Maintains a hypothesis tree for structured exploration
Runs experiments in isolated git worktrees
Performs literature search and novelty checks via alphaXiv
Provides a live dashboard and web interface
Validates results on held-out test splits
Supports multiple model providers via LiteLLM

Use it for

Optimizing machine learning model training
Conducting automated software experiments
Synthesizing data and refining code

Alternatives to Arbor

OpenResearch — Run parallel research workflows locally or on remote compute.
Auto Deep Researcher — Automates experiment operations to reduce manual babysitting of training loops.
AI Scientist v3 — Automates end-to-end scientific research from task descriptions to complete papers.
Paper2Agent — It enables direct interaction with research papers and automatic result reproduction.
Biomni — Automates complex wet-lab experimental design with specialized tools and databases.
DiscoveryRSI — Achieves high actionable-reference rates and supports multi-agent evolutionary research search.
AutoResearch — It provides a stateful, reproducible workflow for automated AI research.
Hypoarena — Provides a fully offline environment to test scientific discovery pipeline logic.
DocTree — Generates source-grounded learning articles adapted to your expertise and time
Lune — Accesses trusted knowledge from top conference papers for AI agents.
Scite — It uses direct publisher agreements to verify scientific claims with exact source links.
Jevon Tane — It provides a rigorous, pre-registered framework for AI agents to conduct science together.

All Arbor alternatives