# RUC-NLPIR/Arbor

A generalist autonomous research agent — runs experiments, researches, and iteratively optimizes, autonomously.

Repository: https://github.com/RUC-NLPIR/Arbor
Canonical: https://ross.abutalabs.com/products/ruc-nlpir-arbor
Homepage: https://ruc-nlpir.github.io/Arbor/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: agents, autonomous-agents, autoresearch
Last push: 2026-08-30T12:35:31+00:00

## Health v2 (maintenance only)
Score: 78/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 100, release rhythm 92, longevity 6
- inputs: {"age_days": 94, "days_push": 3, "days_rel": 52, "gap_med": 5, "n_releases_24m": 4}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1047, forks 124 (observed 2026-09-01T02:13:57.653270+00:00)

## What it is
Arbor is a generalist autonomous research agent that takes a benchmark and a goal, then proposes hypotheses, edits code, and runs real experiments while growing a hypothesis tree that preserves failed directions. It ships as a pip-installable Python CLI (`pip install arbor-agent`) and reports beating Claude Code and Codex by 2.5x on matched compute across tasks like BrowseComp, Terminal-Bench, and MLE-Bench.

## Use cases
- run autonomous ML experiments that iteratively optimize a benchmark score
- have an agent research, edit code, and test hypotheses without supervision
- automate machine-learning engineering tasks like model tuning and data fixes
- keep a record of failed experiment directions instead of forgetting them
- evaluate an autonomous agent on terminal and web-research benchmarks
- reproduce a published autonomous-research agent from an arXiv paper

## When to choose
- you want a hands-off agent that runs and iterates on experiments against a measurable benchmark
- you need optimization loops that only keep gains verified on held-out data
- you want to study or compare against state-of-the-art autonomous research agents

## When to avoid
- you need a tightly supervised, human-in-the-loop coding assistant
- you lack the LLM API access or compute budget for long autonomous experiment runs
- you want a deterministic scripted pipeline rather than an exploratory agent

## Facets
- artifact type: cli-tool
- maturity: active
- function: agent-framework, workflow-automation, machine-learning, developer-tools
- domain: artificial-intelligence, large-language-models, machine-learning, developer-tools
- platform: python, cli, cross-platform
- tags: autonomous-agent, hypothesis-tree, autoresearch, experiment-automation, ml-engineering, benchmark-optimization, tree-search, research-paper, ai-agents, automation

## Member repositories
- RUC-NLPIR/Arbor (main) score 78

## Provenance
- Observed fields: from GitHub, fetched 2026-09-01T02:13:57.653270+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:02:52.270225+00:00, confidence not recorded.
  - readme: https://github.com/RUC-NLPIR/Arbor (fetched 2026-09-01T02:13:57.653270+00:00, sha 0ea4a7bed8dd)
  - homepage: https://ruc-nlpir.github.io/Arbor/ (fetched 2026-08-29T13:04:09.742780+00:00, sha a4875ea1bfd9)
- Data as of 2026-08-30T08:39:29.467469+00:00.