OpenResearch

alphaXiv · Autonomous & Async Agents · Updated Sep 16, 2026

OpenResearch from alphaXiv is a local-first workspace that turns Claude Code, Codex, OpenCode, or Cursor into research agents that review literature, form hypotheses, edit code, launch experiments, inspect results, and produce research artifacts. `orx up` starts a local dashboard on 127.0.0.1:4791 backed by SQLite; each research direction gets its own agent session and isolated git worktree, and variants are tracked in a git-native experiment tree where every run receives an immutable archive of its recorded commit with logs, diffs, and artifacts tied to it. In autoresearch mode, agents run the full propose → change → launch → inspect → decide loop autonomously, in parallel. The same committed snapshot can execute locally, over SSH (`orx up --remote user@host`), or on Slurm, Kubernetes, Ray, Hugging Face Jobs, Modal, Tinker, and managed compute, and `orx install-skills` adds an OpenResearch skill to supported coding agents along with `orx discover` and `orx paper` for literature lookup.

Recent Updates

~4.1k GitHub stars and 340+ commits as of September 2026; reached #1 on GitHub Trending for the day and added Cursor as a supported agent and a Windows beta in September 2026.

As of Sep 16, 2026

Key Features

Turns Claude Code, Codex, OpenCode, or Cursor into research agents with harness and model chosen per session
Parallel agent sessions in isolated git worktrees with a git-native experiment tree
Autonomous autoresearch loop: propose, change code, run, inspect evidence, iterate
Run the same snapshot locally, over SSH, or on Slurm/Kubernetes/Ray/Modal/HF Jobs/Tinker
Local-first SQLite store with projects, runs, logs, and artifacts kept on your machine
`orx` CLI plus installable agent skill, arXiv/DOI paper lookup, and local-model support via LM Studio, Ollama, or oMLX

Who It's For

✓ Best For

ML researchers and engineers who want coding agents to run reproducible, parallel experiments with lineage tracking instead of ad-hoc notebooks and scratch branches.

✗ Not Ideal For

General app development or teams that need multi-user auth on the workspace (the remote service binds to loopback with no application-level authentication). Official builds send opt-out coarse usage analytics.

Pricing

Open Source

Free and open-source under MIT; install with `curl -LsSf https://openresearch.sh/install.sh | sh` (macOS/Linux) or the Windows beta download. Runs on your own hardware or remote hosts; an openresearch.sh account is only needed for optional managed OpenResearch compute and organizations.

Verdict

A well-scoped bridge between coding agents and rigorous experiment tracking; the git-native experiment tree and multi-backend execution make autoresearch practical beyond a laptop.

Track your AI stack