advaitpaliwal/feynman

What it solves

Feynman is an open-source AI research agent designed to automate the process of gathering evidence, analyzing scientific literature, and synthesizing findings. It solves the friction of manual research by providing tools to search papers, rank literature based on evidence, audit claims against codebases, and generate structured research briefs or drafts.

How it works

Built on the Pi agent runtime, Feynman utilizes a multi-agent architecture consisting of a Researcher, Reviewer, Writer, and Verifier. It integrates with alphaXiv for paper search and analysis, the Hugging Face Hub for dataset and model metadata, and a wide array of specialized scientific connectors (Feynman Bio Tools) for domains like biology, chemistry, and genetics. The system is source-grounded, meaning all research claims are linked to papers, documents, or repositories via direct URLs.

Who it’s for

It is designed for AI researchers, scientists, and developers who need to conduct deep literature reviews, replicate ML experiments, find implementable training recipes, or monitor specific research topics.

Highlights

• Deep Research Workflows: Includes multi-agent investigations (/deepresearch), literature reviews (/lit), and the ability to rank papers based on citations, methods, and reproducibility (feynman rank).
• Science Workbench: A standalone local workbench (feynman serve) that provides project navigation, artifact previews for science-specific formats (like LaTeX, chemistry sketches, and genomes), and compute management.
• Extensive Bio-Science Integration: Native connectors for PubMed, ChEMBL, PubChem, UniProt, and dozens of other specialized biological and chemical databases.
• ** uma-agentic Capabilities**: Can plan replication checks, audit paper claims against public codebases, and find ranked ML training recipes from papers and docs.