A personal research assistant, ScienceClaw, built on LangChain DeepAgents and AIO Sandbox, addresses the need for a secure, transparent, and user-friendly research environment. It targets researchers, developers, and students, offering a powerful platform with over 1,900 scientific tools that allows immediate focus on research tasks without complex setup.

How It Works

ScienceClaw operates entirely within Docker containers, ensuring a security-first approach by isolating the agent from the host system and executing code in a sandboxed environment. All generated data remains local within a ./workspace directory, guaranteeing privacy. The architecture provides full transparency, making every step of the agent's workflow—from web searches and data crawling to reasoning and report generation—visible and traceable, allowing users to understand how conclusions are reached. This design prioritizes ease of use, shipping with curated tools and skills for immediate deployment.

Quick Start & Requirements

• Windows Users: A desktop application offers a one-click install, eliminating the need for Docker or command-line interaction. Download the installer from ScienceClaw Desktop v0.0.4 (.tar.gz).
• macOS/Linux Users: Requires Docker and Docker Compose. Recommended system RAM is 8 GB or more. Installation involves cloning the repository (git clone https://github.com/AgentTeam-TaichuAI/ScienceClaw.git), navigating into the directory, and running docker compose -f docker-compose-release.yml up -d --pull always. The application is accessible via http://localhost:5173 with default credentials admin/admin123.
• Developers: Building from source requires docker compose up -d --build.
• Demo: A demo is mentioned, but no direct link is provided in the README.

Highlighted Details

• Integrates ToolUniverse, providing access to over 1,900 scientific tools across diverse domains like drug discovery, astronomy, earth science, chemistry, and academic literature.
• Supports multi-format content generation, producing professional reports in PDF, DOCX, PPTX, and XLSX formats with features like cover pages, tables of contents, charts, and academic styling.
• Operates fully locally and privacy-first, with all data confined to the user's machine.
• Features an extensible skill system and custom tool creation via natural language or Python code, alongside practical features like Feishu (Lark) integration, scheduled tasks, and a built-in file management system.

Maintenance & Community

The project is developed by Zhongke Zidong Taichu (Beijing) Technology Co., Ltd., with notable contributors including Zhiyuan Li and Guangchuan Guo. Community engagement is encouraged through GitHub Issues for feedback and contributions. Version v0.0.1 was released on March 13, 2026.

Licensing & Compatibility

ScienceClaw is released under the MIT License, which is permissive and generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The README does not explicitly state any limitations or known bugs. The project's recent v0.0.1 release suggests it may still be in early development stages.