Feynman — The Open Source AI Research Agent

A research-first CLI agent that searches papers, synthesizes evidence, replicates experiments, and delivers source-grounded research briefs.

Feynman is an open source AI research agent that runs in your terminal. Point it at a research question and it searches academic papers, crawls the web, dispatches parallel subagents, verifies citations, and delivers structured research briefs with inline sources — all from a single command.

Installation

Install Feynman on macOS, Linux, or Windows in one command.

Quick Start

Run your first deep research session in under five minutes.

Research Workflows

Explore deep research, literature review, replication, audit, and more.

CLI Reference

Full reference for every command, flag, and slash workflow.

What Feynman does

# Ask a research question naturally
feynman "what do we know about scaling laws in LLMs?"

# Run a source-heavy multi-agent investigation
feynman deepresearch "mechanistic interpretability"

# Literature review from primary sources
feynman lit "RLHF alternatives"

# Compare a paper's claims against its public codebase
feynman audit 2401.12345

# Replicate an experiment on local or cloud GPUs
feynman replicate "chain-of-thought improves math"

How it works

Feynman orchestrates four specialized subagents — Researcher, Writer, Reviewer, and Verifier — over a pool of tools: AlphaXiv for paper search and Q&A, web search for current information, Docker and cloud GPU compute for experiment execution, and session memory for recall across sessions. Every output links claims to papers, docs, or repositories with direct URLs.

1. Install

   Run the one-line installer. Feynman ships as a standalone bundle with no external runtime dependencies.

   curl -fsSL https://feynman.is/install | bash
   
2. Set up your model

   Run feynman setup to pick a model provider and authenticate. Feynman supports Anthropic, OpenAI, Google, OpenRouter, and more.

   feynman setup
   
3. Start researching

   Launch the interactive REPL or pass a question directly. Use slash commands to invoke specific research workflows.

   feynman
   > /deepresearch transformer architectures for protein folding
   

Key capabilities

Deep Research

Multi-agent investigation with parallel researchers, synthesis, and a verification pass. Produces a cited brief saved to outputs/.

Literature Review

Structured review from paper search and primary sources, with consensus, disagreements, and open questions.

Experiment Replication

Replicate paper results locally, in Docker, or on Modal / RunPod GPU infrastructure.

Paper Audit

Compares a paper’s claims against the public codebase to surface mismatches.

Peer Review Simulation

Severity-graded feedback on drafts or papers, with a concrete revision plan.

AlphaXiv Integration

Search, read, Q&A, and annotate academic papers via the alpha CLI.