AutoResearch
Open-source agent workflow that runs AI/ML research from idea to paper-ready evidence
AutoResearch is an open-source agent workflow that takes a research idea, or one it discovers from papers and developer communities, through experiment planning, implementation, execution, and independent review, producing an evidence package for paper writing. It persists plans, code, logs, metrics, and critic reports to disk so a run can be inspected, taken over, or resumed after interruption, and it preserves negative results instead of only showing successes. The project states plainly that it 'cannot guarantee every conclusion is correct.'
What holds up
- +Preserves negative results and stops on a failed hypothesis instead of forcing a success narrative
- +Runs are stateful and resumable — plans, code, logs, and critic reports persist to disk for review
- +Uses at least three distinct models for independent review, so one model can't approve its own work
Mind the limits
- −States outright it can't guarantee a conclusion is correct — findings still need researcher review
- −Runs on GPT, Gemini, and Claude under the hood, not its own engine — you need API access to those
Featured here? Take the badge
Put it on your site — it links back to this review. Free for every listed product, always.
<a href="https://stillworks.dev/products/p/autoresearch/"><img src="https://stillworks.dev/badge/autoresearch.svg" alt="Picked by StillWorks" width="250" height="54"></a>