ScienceClaw is a personal research assistant built with LangChain DeepAgents and AIO Sandbox infrastructure, adopting a completely new architecture beyond OpenClaw. It offers stronger security, better transparency, and a more user-friendly experience.

• [2026-03-13] ScienceClaw v0.0.1 is officially released! Visit our website: scienceclaw.taichuai.cn

No Docker required, no command-line needed. Download the desktop installer and get started instantly.

1. Download the installer

👉 ScienceClaw Desktop v0.0.4 (.tar.gz)

2. Extract and install

After downloading, extract the archive and run the installer. Follow the setup wizard to complete installation.

3. Launch

Once installed, double-click the desktop shortcut to start ScienceClaw — ready to use out of the box.

• Docker & Docker Compose (Docker Desktop includes Compose)
• Recommended system RAM ≥ 8 GB

1. Get the code

• Fresh install:

git clone https://github.com/AgentTeam-TaichuAI/ScienceClaw.git
cd ScienceClaw

• Upgrade an existing installation:

cd ScienceClaw
git pull

2. Launch — pull pre-built images

docker compose -f docker-compose-release.yml up -d --pull always

Pulls pre-built images directly — no local compilation needed. Ready in a few minutes.

3. Open in browser

http://localhost:5173

4. Login

Default admin credentials: Field Value Username admin Password admin123 ⚠️ Please change the default password after your first login.

docker compose up -d --build

Builds all images from source code. Ideal for developers who need to modify the code. The first build downloads dependencies and may take longer.

To lower the barrier for new users, a limited batch of LLM API resources:

Limited availability — first come, first served. We will continue to secure more compute resources for the community.

ScienceClaw integrates ToolUniverse, a unified ecosystem of 1,900+ scientific tools spanning multiple disciplines:

ScienceClaw makes it easy to extend with your own tools:

• Natural language creation — Simply describe what you want in chat, and the agent will create, test, and save a new tool for you automatically.
• Manual mounting — Drop any Python file with @tool decorated functions into the Tools/ directory; they are auto-detected and hot-loaded without restart.

Skills are structured instruction documents (SKILL.md) that guide the Agent through complex, multi-step workflows. Unlike tools (executable code), skills act as the Agent's "playbook" — defining strategies, rules, and best practices.

ScienceClaw can produce professional research deliverables in 4 document formats:

• Natural language creation — Describe your workflow in chat, and the agent will draft, test, and save a new skill for you.
• Manual installation — Place a folder with a SKILL.md file into the Skills/ directory. The agent automatically matches and loads relevant skills based on user intent.
• Community ecosystem — Discover and install skills from the open community with the built-in find-skills capability.

ScienceClaw/
├── docker-compose.yml              # 10-service orchestration
├── docker-compose-release.yml      # Pre-built image orchestration (for end users)
├── docker-compose-china.yml        # China mirror acceleration
├── images/                         # Static assets (logo, screenshots)
├── videos/                         # Demo videos
├── Tools/                          # Custom tools (hot-reload)
├── Skills/                         # User & community skill packages
├── workspace/                      # 🔒 Local workspace (data never leaves your machine)
└── ScienceClaw/
    ├── backend/                    # FastAPI backend
    │   ├── deepagent/              # Core AI agent engine (LangGraph)
    │   ├── builtin_skills/         # 9 built-in skills (pdf, docx, pptx, xlsx, tooluniverse, ...)
    │   ├── route/                  # REST API routes
    │   ├── im/                     # IM integrations (Feishu / Lark)
    │   ├── mongodb/                # Database access layer
    │   ├── user/                   # User management
    │   ├── scripts/                # Utility scripts (Feishu setup, etc.)
    │   └── translations/           # i18n language packs
    ├── frontend/                   # Vue 3 + Tailwind frontend
    ├── sandbox/                    # Isolated code execution environment
    ├── task-service/               # Scheduled task service (cron jobs)
    └── websearch/                  # Search & crawl microservice

# First launch (recommended for most users) — pull pre-built images, no local compilation
docker compose -f docker-compose-release.yaml up -d

# First launch (for developers) — build from source and start all services
docker compose up -d --build

# First launch (for developers in China) — build with Chinese mirror sources for faster downloads
docker compose -f docker-compose-china.yml up -d --build

# Daily launch — fast startup, no rebuild needed
docker compose up -d

# Check service status
docker compose ps

# View logs (-f to follow in real time)
docker compose logs -f backend     # Backend logs
docker compose logs -f frontend    # Frontend logs
docker compose logs -f sandbox     # Sandbox logs

# Restart a single service
docker compose restart backend

# Stop all services
docker compose down

# Stop a single service
docker compose stop backend

ScienceClaw is built entirely on Docker, so uninstalling it is clean and simple — it has no side effects on your host system.

# Stop and remove all containers
docker compose down

# (Optional) Remove downloaded images to free up disk space
docker compose down --rmi all --volumes

Then simply delete the project folder:

rm -rf /path/to/ScienceClaw

That's it. No residual files, no registry entries, no system-level changes.

Zhongke Zidong Taichu (Beijing) Technology Co., Ltd.

We welcome contributions, feedback, and discussions! Join our community:

• Submit issues and feature requests via GitHub Issues
• Share your custom tools and skills with the community

MIT License

ScienceClaw is built on the shoulders of excellent open-source projects. We would like to express our sincere gratitude to:

• LangChain DeepAgents — The batteries-included agent harness built on LangChain and LangGraph. ScienceClaw's core agent engine is powered by the DeepAgents architecture, which provides planning, filesystem access, sub-agent delegation, and smart context management out of the box.

• AIO Sandbox — The all-in-one agent sandbox environment that combines Browser, Shell, File, and MCP operations in a single Docker container. ScienceClaw relies on AIO Sandbox to provide secure, isolated code execution with a unified file system.

• ToolUniverse — A unified ecosystem of 1,900+ scientific tools developed by the Zitnik Lab at Harvard. ToolUniverse powers ScienceClaw's multi-disciplinary research capabilities across drug discovery, genomics, astronomy, earth science, and more.

• SearXNG — A privacy-respecting, hackable metasearch engine. ScienceClaw uses SearXNG as the backbone of its web_search tool, aggregating results from multiple search engines without tracking.

• Crawl4AI — An open-source, LLM-friendly web crawler. ScienceClaw's web_crawl tool is powered by Crawl4AI, enabling intelligent content extraction from web pages for research and analysis.

• Zhiyuan Li
• Guangchuan Guo
• Shaoling Lin
• Songsong Lei
• Zhidong Zhang

Technical Support: NLP Group, Institute of Automation, Chinese Academy of Sciences