What is Feynman?
Feynman is an open-source AI research agent made by Companion. It reads papers, searches the web, writes research drafts, plans experiments and cites research claims. The project is released under the MIT licence, and the code is on GitHub.
You run it from a terminal. Workflows are available as slash commands in the REPL or as feynman commands from the shell: deep research, literature review, internal review, paper-to-code audit, replication plan, ranked ML training recipes, source comparison, paper-style drafts and a bounded autoresearch loop. Results are saved to an outputs folder, and some workflows also write a provenance file listing the sources used and what was verified.
Feynman searches alphaXiv, Semantic Scholar, OpenAlex, arXiv, PubMed, Europe PMC, bioRxiv, medRxiv, Crossref, the web and Hugging Face, and reads local PDFs and documents. Installation uses a curl or PowerShell script that installs a standalone bundle, or npm with Node.js 22.22 or newer. After installing you sign in to a model provider with feynman setup. The README states that Feynman sends anonymous usage data to PostHog and that it can be turned off.
Key capabilities
- Ten workflows, including /deepresearch, /lit, /review, /audit, /replicate, /recipe, /compare, /draft, /autoresearch and /summarize.
- Search across alphaXiv, Semantic Scholar, OpenAlex, arXiv, PubMed, Europe PMC, bioRxiv, medRxiv, Crossref and Hugging Face.
- Provenance files that list the sources used and what was verified.
- Install scripts for macOS, Linux and Windows, plus an npm package.
- A skills-only bundle for Codex and OpenCode.
Where Feynman fits
Feynman fits into literature discovery, manuscript review and early experimental planning. A research team can use it to compare sources, examine whether public code supports a paper's claims, or prepare a replication plan. Its replication workflow waits until an execution environment has been chosen before running.
Before deployment, teams should verify model configuration, access to scholarly services and the environment intended for experiments. OpenAlex and Semantic Scholar keys can avoid shared rate limits. Teams should also review telemetry settings: Feynman sends usage and error information to PostHog, but excludes prompts, model output, paper content and tool arguments. Telemetry can be disabled through configuration.
Install
npm install -g @companion-ai/feynmanReviews of Feynman
Be the first to share your experience with Feynman.
Write a review
Loading review form…