Overview

Feynman is an open-source AI research agent designed to accelerate scientific discovery and knowledge synthesis. It empowers researchers, developers, and curious minds to interact with complex information, conduct deep investigations, and generate auditable research outputs. Built on the robust Pi agent runtime and leveraging tools like AlphaXiv, Feynman transforms how we approach research by automating information retrieval, analysis, and synthesis.

Key Features

• Effortless Installation: Simple one-line installers for macOS, Linux, and Windows, with an npm alternative for existing Node.js environments.
• Powerful Research Workflows: Execute sophisticated research tasks directly from the command line, including literature reviews, deep research investigations, paper analysis, citation ranking, and replication planning.
• Integrated Science Workbench: A comprehensive local development environment for managing projects, chatting with AI assistants, previewing artifacts, managing compute resources, and more.
• Extensive Skill & Toolset: Seamless integration with AlphaXiv for paper search and Q&A, Hugging Face Hub for model and dataset inspection, and a vast array of specialized Bio Tools for scientific literature and data.
• Auditable Outputs: Ensures that research claims are grounded in verifiable sources, with direct links to papers, documentation, or code repositories.
• Multi-Agent Capabilities: Employs specialized agents (Researcher, Reviewer, Writer, Verifier) for decomposition and targeted task execution.
• Local Model Support: Easily integrates with popular local LLM setups like LM Studio, LiteLLM, Ollama, and vLLM.

Typical Use Cases

• Accelerated Literature Review: Quickly synthesize information from papers and web sources into cited research briefs with feynman "what do we know about...".
• Deep Investigation: Conduct multi-agent, source-heavy investigations with commands like feynman deepresearch "mechanistic interpretability".
• Informed Reading Decisions: Prioritize research papers with feynman rank "topic" based on citation, method, reproducibility, and provenance evidence.
• Reproducible Science: Plan and execute replication checks for research papers using feynman replicate "paper title or ID".
• ML Recipe Discovery: Find and rank implementable ML training recipes from papers, datasets, and code with feynman recipe "task".
• Comprehensive Research Environment: Utilize the standalone feynman serve command to access a local workbench for project management, AI-powered chat, and artifact previewing.
• Knowledge Management: Automatically update and maintain a local citation graph with feynman rank "topic" --expand-citations.