feynman

Feynman is an open-source AI research agent designed to assist with scientific literature research and analysis. It operates primarily as a command-line tool (`feynman`) and also offers a standalone science workbench launched via `feynman serve`. The tool supports a range of workflows including deep research, literature reviews, paper ranking, paper access resolution, replication planning, and ML recipe discovery. Key capabilities include: - **Research workflows**: Commands like `deepresearch`, `lit`, `rank`, and `paper` enable multi-agent investigations, literature reviews with consensus and open questions, paper scoring based on citations and reproducibility, and full-text fetching. - **Agent system**: Four bundled agents (Researcher, Reviewer, Writer, Verifier) handle evidence gathering, critique, drafting, and citation verification. - **Integration**: Leverages alphaXiv for paper search and Q&A, Hugging Face Hub for dataset/model inspection, and supports multiple LLM providers (including local models via LM Studio, LiteLLM, Ollama, or vLLM). - **Science workbench**: `feynman serve` opens a web-based interface with projects, chat, notebooks, artifact previews (including chemistry sketches, PDFs, notebooks), and provenance tracking. - **Installation options**: Standalone native bundles for macOS/Linux/Windows, an npm package, and a skills-only installer for Codex, Claude, or OpenCode. Feynman is built on the Pi agent runtime and is MIT-licensed. It provides extensive tooling for scientific discovery, including access to databases like PubMed, ChEMBL, Ensembl, and many others through its Feynman Bio Tools, making it suitable for researchers in life sciences and beyond.