Feynman
An open-source command-line AI agent that helps with literature reviews and research.
- Stars
- 10.1k
- Forks
- 1.1k
- Licence
- MIT
- Language
- TypeScript
- Last update
- 6 days ago
- Type
- AI model / agent
What it is
Feynman is a command-line tool that runs an AI research agent. You give it a topic or a paper, and it searches academic sources such as arXiv, Semantic Scholar, and PubMed, then produces things like literature reviews, critiques, and drafts. It reads local PDFs too and saves results to an outputs folder, with a provenance file listing the sources used.
- Run a literature review with feynman lit "RLHF alternatives"
- Get a planned multi-source investigation with feynman deepresearch "mechanistic interpretability"
- Critique a draft paper with /review and get severity levels and a revision plan
- Compare claims in a paper against its public code with /audit
Great for
- Students starting a literature review on an unfamiliar topic
- Researchers who want a first pass at related work and open questions
- People comfortable running terminal tools who want to check paper claims against public code
- Anyone who wants saved, source-linked research notes rather than a chat answer
Not ideal for
- People who need guaranteed-accurate citations without checking them, since the README describes verification steps but does not promise correctness
- Users who do not want usage data sent elsewhere by default, because telemetry is on unless disabled
- Anyone who needs a graphical app; the README describes a CLI and does not mention a GUI
How to get it
- 1
Install with a package manager
You'll need: Node.js 22.22 or newer (for the npm install route) · A model provider account for feynman setup
Install with npm (requires Node.js 22.22+)
npm install -g @companion-ai/feynmanSign in to a model provider
feynman setup - 2
Run it
- Install Feynman using the curl script or npm command
- Run feynman setup and sign in to a model provider
- Run feynman lit "your topic" for a literature review
- Check the outputs/ folder for saved results and the provenance file
- 3
Get the most out of it
- Add free OPENALEX_API_KEY and SEMANTIC_SCHOLAR_API_KEY keys to avoid shared rate limits
- Use /deepresearch for planned investigations with parallel researchers and citation checks
- Use /summarize on a paper, repository, or PDF for a quick overview
- Read the setup guide for local models and editor integration such as Zed
Downloads
Strengths
- Open source under the MIT license, with a public repository and 10,040 stars
- Searches many academic sources and saves outputs with a provenance file
- Installs with a single command on macOS, Linux, and Windows, or through npm
Trade-offs
- Telemetry is on by default and sends commands, errors, and stack traces; it can be turned off with FEYNMAN_TELEMETRY=off
- Requires a sign-in to a model provider, so you may need an account and pay for model usage (the README does not detail costs)
- The README is truncated and does not explain the agent's accuracy limits in depth
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| feynman (this) | 10.1k | AI model / agent | intermediate | 6 days ago | Students starting a literature review on an unfamiliar topic |
| last30days-skill | 66.3k | AI model / agent | beginner | 5 days ago | People who want a recent, community-driven picture of a topic from the past 30 days |
| deep-research | 19.9k | AI model / agent | intermediate | 7 months ago | Developers who want a readable example of a deep research agent to learn from or extend |
| hermes-agent | 260k | AI model / agent | intermediate | 4 days ago | Developers who want a personal agent they can host on their own server or machine |
| learn-claude-code | 79.7k | Learning resource | intermediate | 10 days ago | Developers who want to understand how coding agents like Claude Code are structured |
| career-ops | 79.4k | AI model / agent | intermediate | 4 days ago | Job seekers who apply to many listings and want to filter them before spending time on each one |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.