Get a cited research summary on any ML topic by typing a question into your terminal.
Generate a literature review that maps where academic sources agree and disagree on a subject.
Audit a research paper by checking whether its public codebase actually matches the claims it makes.
Reproduce paper experiments locally or on cloud GPU services with a single /replicate command.
Basic queries work without API keys using local models via Ollama or LM Studio, cloud GPU features require Modal or RunPod accounts.
Feynman is an open-source command-line tool that acts as a research agent for scientific and machine learning topics. You give it a question or a research topic, and it searches through academic papers, web sources, and code repositories to produce a cited, source-grounded brief in return. It is designed primarily for people working in machine learning and AI research who want to move quickly from a question to a structured answer. The tool runs in your terminal and accepts both plain English questions and slash commands for specific workflows. Asking "what do we know about scaling laws" returns a research summary with citations. Running the /lit command on a topic produces a literature review that notes where sources agree and disagree. The /audit command takes a paper ID and checks whether the claims in the paper match the public codebase it references. The /replicate command attempts to reproduce experiments from a paper on your local machine or in cloud GPU environments. Four built-in agents work together behind the scenes. A Researcher gathers evidence from papers and documentation. A Reviewer applies simulated peer-review feedback with severity grades. A Writer produces structured drafts from collected notes. A Verifier checks inline citations, confirms source URLs, and removes broken links from the output. External integrations extend what the tool can reach. It connects to AlphaXiv for paper search and Q&A, the Hugging Face Hub for dataset and model inspection, Docker for isolated experiment execution, and services like Modal and RunPod for GPU compute when an experiment needs more resources than a local machine can supply. Installation is a single curl or PowerShell command on macOS, Linux, or Windows. If you only want the research skill library without the full terminal application, a separate installer is available for that subset. The project also supports local AI models through LM Studio, Ollama, or a LiteLLM proxy, so you do not need cloud API keys to run basic queries.
← companion-inc on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.