Feynman: New Open-Source AI Research Agent Enables Local Paper Reading, Web Search, and Experiment Running

Summary

Feynman is an open-source AI research agent that automates key aspects of academic research workflows. The tool can read scientific papers, search the web, write drafts, run experiments, and crucially, cite every claim it makes—all running locally on users' computers. Designed to augment researcher productivity, Feynman supports both slash commands and natural language interfaces, allowing researchers to ask questions or run workflows with every answer grounded in cited sources.

The agent features a modular architecture that assembles the right set of capabilities for each task, built on the Pi framework and integrated with arXiv for paper access. Users can execute experiments locally or scale to managed GPU infrastructure when computational demands require it. The project emphasizes source-grounding throughout, with capabilities shipped as composable Pi skills and all outputs maintaining traceability to their sources.

• Addresses a critical gap in AI research tools by ensuring every claim is cited, supporting academic rigor and reproducibility

Editorial Opinion

Feynman represents a thoughtful approach to AI-assisted research by prioritizing source attribution and local control—two aspects often compromised in mainstream AI tools. By building on established open standards and keeping computations local by default, the project respects both researcher autonomy and the academic integrity requirements of scholarly work. This could become a template for how research-focused AI tools should handle citations and transparency.