This project provides an autonomous AI research assistant designed to streamline the scientific discovery process. It targets researchers and power users in AI/ML, automating tasks from literature review and hypothesis generation to formal theorem proving and paper drafting, thereby accelerating the path to publishable results.

How It Works

EurekaClaw operates as a multi-agent system, autonomously navigating the research lifecycle. It begins by crawling and synthesizing information from academic sources like arXiv and Semantic Scholar. The system then generates novel hypotheses, formalizes and verifies mathematical proofs through a detailed 7-stage pipeline, and drafts camera-ready LaTeX papers. Its local-first, privacy-by-design architecture ensures compatibility with various model APIs while keeping user data secure. Continual learning allows the AI to distill and improve its proof strategies over time.

Quick Start & Requirements

Installation for macOS/Linux is facilitated by a shell script: curl -fsSL https://eurekaclaw.ai/install.sh | bash. Windows users are directed to use WSL 2 and follow the macOS/Linux instructions. A manual installation requires cloning the repository, ensuring Python ≥ 3.11, Node.js ≥ 20, and Git are installed, then running make install. Post-installation, eurekaclaw onboard configures API keys (Anthropic Claude recommended, with OAuth option) and settings, followed by eurekaclaw install-skills to set up proof capabilities. Detailed documentation is available at https://eurekaclaw.github.io/.

Highlighted Details

• Autonomous Research Pipeline: Integrates literature crawling, hypothesis generation, formal theorem proving, and paper writing into a single workflow.
• Formal Theorem Prover: Employs a 7-stage bottom-up pipeline for generating, verifying, and formalizing proofs, leveraging tools like Lean4.
• Browser UI: Features a React + TypeScript interface for live agent tracking, proof sketching, and skill management.
• Continual Learning: Distills proof strategies into reusable skills after each session, enhancing future performance.
• Scientist-Bench Evaluator: Includes a framework for evaluating sessions based on formal correctness, novelty, experimental alignment, and citation coverage.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels such as Discord or Slack.

Licensing & Compatibility

EurekaClaw is released under the permissive Apache 2.0 License, allowing for broad compatibility with commercial use and closed-source integration. Its local-first and privacy-by-design principles further enhance its suitability for sensitive research environments.

Limitations & Caveats

Native Windows support is currently under active development and not fully supported. The Experiment Runner feature, intended for numerical validation of theoretical bounds, is also still under development.