feynman: Research Papers with an AI Agent

Overview

Feynman is a command-line research agent that helps turn a topic, paper, or research question into structured findings. It searches academic and web sources, then uses a workflow of researcher, verifier, reviewer, and writer agents to produce research outputs with source provenance for selected workflows.

It is aimed at researchers, students, and technical teams who need more than a quick paper summary: for example, a literature review, a comparison of claims, or a critique with a revision plan. It runs on Pi and requires signing in to a model provider.

Key Features

• Investigates topics using planned research and parallel researchers, with citation checks.
• Produces literature reviews covering consensus, disagreements, and open questions.
• Reviews papers or drafts with severity levels and a revision plan.
• Compares paper claims against public code, or creates a replication plan that runs only after an environment is chosen.
• Builds source comparison matrices, ML training recipes, and paper-style drafts.
• Summarizes papers, repositories, and PDFs, and can read local documents.
• Searches academic indexes, the web, and Hugging Face; selected workflows save provenance files listing sources and verification status.

Use Cases

• Researchers mapping agreement and unresolved questions across papers before starting a project.
• Students preparing a literature review and wanting citations and source provenance to inspect.
• Authors seeking structured feedback on a draft and a prioritized revision plan.
• ML practitioners comparing a paper's claims with its public implementation or planning a replication.
• Developers exploring a technical topic who want research findings saved as reusable files.

Getting Started

Install with npm using Node.js 22.22 or newer, then configure a model provider:

npm install -g @companion-ai/feynman
feynman setup

See the README and documentation for other installation options and usage details.

Alternatives

• Biomni: Biomni is a biomedical agent for executing diverse research tasks, while Feynman focuses on investigating topics, reviewing papers, and comparing sources.

Considerations

• Feynman is a model-backed agent, so you must configure a model provider; local models are also covered in the setup guide.
• Research results depend on the sources retrieved and should be checked against the cited material. The project describes citation checks and provenance for some workflows, not a guarantee that every conclusion is correct.
• Shared API rate limits may apply to source services. The README notes that free OpenAlex and Semantic Scholar API keys can avoid shared limits.
• Anonymous usage telemetry is enabled by default and includes command and workflow data, token counts, and errors with messages and stack traces. The project says it does not send prompts, model output, paper content, or tool arguments; telemetry can be disabled with FEYNMAN_TELEMETRY=off.