A self-evolving AI research colleague, ScienceClaw addresses the limitations of general-purpose AI assistants for scientific endeavors. It targets researchers by providing specialized tools for literature searching, database querying, and analysis, aiming to improve accuracy and efficiency through continuous learning and persistent memory.

How It Works

ScienceClaw's core innovation lies in its self-evolving skills, which adapt to a user's specific research domain and workflow over time, unlike static AI capabilities. It features persistent research memory, storing context and patterns across sessions using temporal decay weighting and LanceDB for efficient retrieval. Engineered for long-duration research, it supports sessions exceeding an hour, enforcing depth through multi-phase protocols and mandatory tool-call thresholds to prevent shallow conclusions. A paramount feature is its "Zero Hallucination" protocol, strictly ensuring all citations originate from verified tool results within the current conversation, enforced by a comprehensive research protocol.

Quick Start & Requirements

• Primary install: Clone the repository (git clone https://github.com/beita6969/ScienceClaw.git), navigate into the directory (cd ScienceClaw), and run the one-click setup script: ./setup.sh. This installs Node, Python, MCP servers, and skills.
• Prerequisites: Node.js, Python, and pnpm are required. The setup.sh script automates most dependency installations.
• Research Features: For manual setup or advanced configuration, edit ~/.openclaw/openclaw.json to enable memory plugins (memory-core, memory-lancedb) and configure agent heartbeat intervals.
• Links: Project repository: https://github.com/beita6969/ScienceClaw.git.

Highlighted Details

• Starts with 285 skills, with a self-evolution system that generates new skills at runtime.
• Integrates with over 25 academic databases across natural and social sciences, including PubMed, UniProt, and World Bank.
• Employs a strict "Zero Hallucination" protocol, governed by a 629-line SCIENCE.md research protocol, ensuring factual accuracy and verifiable citations.
• Designed for extended research sessions (1 hour+), featuring smart context management and mandatory depth thresholds for thorough analysis.

Maintenance & Community

Direct contact is available via email at mingdazhang@ieee.org. The README does not specify community channels like Discord or Slack, nor does it detail active contributors or sponsorships.

Licensing & Compatibility

The project is released under the MIT License, which is generally permissive for commercial use and integration with closed-source projects. No specific compatibility restrictions are noted.

Limitations & Caveats

ScienceClaw is highly specialized for academic research, potentially making it less suitable for general-purpose AI tasks. The self-evolving nature of its skills, while powerful, may introduce an element of unpredictability or require ongoing monitoring. The README does not detail known bugs, alpha status, or specific unsupported platforms.