CoAutoResearch CLI
A research system on top of Codex CLI or Claude Code: runs of experiments, discussion of progress and a paper draft.
- 115 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent runs code and experiments on your machine on its own across long series of trials.
- The work spends subscription limits and compute budget that you should set in the brief.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/coautoresearchDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs code and experiments on your machine on its own across long series of trials.
- The work spends subscription limits and compute budget that you should set in the brief.
This is third-party code. Review the repository files before installing.
What it does
CoAutoResearch starts a local dashboard through which Codex CLI or Claude Code carry out research on your brief. Work is split into Trials: bounded iterations where the agent sets up an experiment while the service checks proposed changes, records accepted results and decides whether to continue, pause or ask a human. In parallel you can discuss ideas in a separate chat and send the agent a suggestion when it is ready. Findings are gathered into a manuscript, and from a frozen evidence snapshot the agent builds a paper draft as a PDF. Internal review stays internal: peer review and the judgement of scientific value fall to a person.
Who it is for. Researchers and engineers who hand series of reproducible experiments to Codex CLI or Claude Code and want to follow progress from a browser.
Good fit when
- You need to compare two methods on a small public dataset under a stated CPU budget.
- There is a proposal and materials, and the agent must find the next useful experiment and prepare a plan for review.
- A result should be tested against a stronger comparison with the original data left intact.
- Experiments run for hours and the outcome should be gathered into a manuscript that links to recorded facts.
Not a fit when
- Independent scientific verification of a claim is required, which internal review does not replace.
- The experiment is unsafe to run unattended, for example when it writes to production databases or uses paid cloud resources.
- A one-off answer on a topic is enough: asking the agent directly is cheaper than trials and a dashboard.
Example request
Compare two methods on a small public dataset. Use CPU only, state the budget and ask before final evaluation.Limitations
Node.js 20 or newer, Python 3.10 or newer, Git and an installed, authenticated Codex CLI or Claude Code are required. PDF output adds Python 3.12 or newer, four pinned scientific skills, LaTeX and Poppler. The user completes the provider login, and model limits depend on their account. The interface and documentation are in English, while agent replies follow the user's language. A created project does not start research by itself: after the brief you press the start button.
How to disable. Stop the `node bin/auto-research.js ui` process and delete the repository directory. The projects folder set with the `--projects-dir` flag is kept separately and is removed by hand.
FAQ
Which agents does CoAutoResearch support?
Codex CLI and Claude Code. Both must be installed and authenticated before launch, and the user performs the provider login.
How do I launch the dashboard manually?
Clone the repository and run `node bin/auto-research.js doctor`, then `node bin/auto-research.js ui`. Open the printed URL and keep the server running.
What is a Trial?
A bounded research iteration. The service checks proposed changes, records accepted results and decides whether to continue, pause or ask a human.
Does creating a project start research right away?
No. After creating a project, send a research brief, review the direction and press start autoresearch.
Can I change the projects folder?
Yes, the `--projects-dir /path/to/projects` flag keeps research outside the repository directory.
Discussion
No questions or reviews yet.
Related
CLI toolsAcademic plugin for Claude Code, Gemini CLI, Codex and OpenCode: literature reviews, peer review and citation checks
A local runtime where Claude Code or Codex run research for hours: idea branches compete for budget and the verified ones win
A Codex skill for an autonomous loop: Claude Code computes, Codex reviews hourly, critiques and sends the next prompt.
Awesome Autoresearch
awesome-autoresearch
A collection of autonomous self-improvement loops and research agents that set hypotheses and test them themselves
A set of 11 Claude Code skills that audits a third-party paper for self-consistency and fabrication and writes a report.
An autonomous research assistant: from idea and plan through experiments and analysis to a paper, relying only on real results
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| CoAutoResearch (this card) | no | 115 | |
| Autoresearch for skills | no | 982 | a skill for autonomous iteration toward a numeric metric; CoAutoResearch adds a dashboard, discussion and a paper draft |
| Codex Autoresearch | no | 3,205 | a loop of measurable repository edits; CoAutoResearch carries scientific research from hypothesis to manuscript |
README in short
Setup can be handed to an agent as a single instruction or done manually from source. Three ways to begin follow: a new question, existing work and a result to investigate. The example papers on digits and the Ising model are described as developer-operated drafts with human intervention rather than external case studies.