ScienceClaw Research Assistant
Run scientific search and analysis in a local workspace, then turn the results into research-ready documents.
- Source repo
- AgentTeam-TaichuAI/ScienceClaw
- Stars
- ★ 679
- Last updated
- 6mo ago
- Primary language
- Python
- FA score
- 36/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Researchers collecting papers from PubMed, arXiv, or OpenAlex can use the built-in academic literature search tools to assemble results.
- Not a fit if
- Teams that need literature search and analysis to work without network access
- Users unwilling to deploy Docker services or use the official cloud
- Source review
- 36/100 · Major gaps
What does this agent do, and when should you use it?
ScienceClaw is a personal research assistant built with LangChain DeepAgents, LangGraph, and AIO Sandbox for literature search, data analysis, and research deliverables. Its Docker Compose deployment includes a Vue 3 web interface, FastAPI backend, scheduled task service, and an isolated code execution environment. The README says it integrates more than 1,900 ToolUniverse scientific tools alongside web search, crawling, file operations, and user-defined Python tools. It can produce PDF, DOCX, PPTX, and XLSX files, stored in a local workspace. The project also offers a Windows desktop installer and an official hosted service; self-hosting requires Docker and Docker Compose.
A user submits a research task through the web interface. ScienceClaw can use web_search and web_crawl to search and retrieve web content, then call listed ToolUniverse capabilities such as PubMed literature search, protein analysis, and drug discovery tools. It can read and write workspace files and run code in the AIO Sandbox, and it can load Python functions decorated with @tool from the Tools/ directory. Built-in Skills guide multi-step work with PDF, DOCX, PPTX, and XLSX files, which the system can generate as research outputs. Users can also schedule tasks and receive results through Feishu notifications.
- Researchers collecting papers from PubMed, arXiv, or OpenAlex can use the built-in academic literature search tools to assemble results.
- Drug discovery and biomedical teams can call listed scientific tools such as OpenTargets, FAERS, UniProt, and PDB.
- Researchers preparing analyses as PDFs, Word documents, or slide decks can use the corresponding built-in Skills.
- Users who need recurring research runs and Feishu group updates can configure scheduled tasks and Feishu notifications.
- Developers who want a research assistant on their own machine and visibility into its work can deploy with Docker Compose and inspect the traceable workflow.
How do you install or deploy this agent?
For self-hosting, install Docker and Docker Compose; the README recommends at least 8 GB of system RAM. Clone the repository and start the prebuilt images:
git clone https://github.com/AgentTeam-TaichuAI/ScienceClaw.git
cd ScienceClaw
docker compose -f docker-compose-release.yml up -d --pull alwaysWhen the services are up, open http://localhost:5173. The documented Windows desktop route is to download the ScienceClaw Desktop v0.0.4 .tar.gz installer, extract it, and run the installer. The README does not give desktop installation steps for other operating systems.
How do you use this agent?
Open http://localhost:5173 and sign in. The README lists the initial administrator credentials as username admin and password admin123, and says to change the password after the first login. Submit a research task in the chat interface. Feishu webhook notifications can be configured in settings, and one-time or recurring runs can be scheduled. Browse or download generated files from the built-in file panel; they are also stored under the local ./workspace directory. To add a tool, place a Python file containing @tool-decorated functions in Tools/. The README does not provide an API invocation example.
What are this agent's strengths and limitations?
- Combines more than 1,900 ToolUniverse scientific tools with literature sources including PubMed, arXiv, and OpenAlex in one research workflow.
- Runs code in the isolated AIO Sandbox and stores generated data in a local workspace.
- Produces PDF, DOCX, PPTX, and XLSX files, with scheduled runs and Feishu notifications.
- Self-hosting requires Docker Compose and multiple services; the README recommends at least 8 GB of RAM.
- Web search, crawling, and external scientific data tools require network access; the README does not document a fully offline mode.
- The README does not identify model providers, API key setup, or pricing, so model costs and provider choice are unclear from the documentation.
- The documented Windows desktop installer is v0.0.4, and desktop installation steps for macOS or Linux are not provided.
How does this agent compare with similar options?
The README describes ScienceClaw as using a new architecture beyond OpenClaw, but gives no feature-by-feature comparison or migration details.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| ScienceClaw Research Assistant This agent | 36 · Major gaps | Self-hosted serviceFreemium | ★ 679 | 6mo ago | Python | — |
| Scientific Agent Skills | 51 · Major gaps | CLIFree + model costs | ★ 56k | 3d ago | Python | Claude Code |
| K-Dense BYOK | 49 · Major gaps | Desktop appFree + model costs | ★ 1.4k | 20d ago | TypeScript | ChatGPT · Codex · Claude.ai |
| Qinyan Academic Skills | 48 · Major gaps | CLIFree + model costs | ★ 1k | 3mo ago | Python | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README claims Docker isolation, a local workspace, and visible workflows, while listing code execution, shell, file writing, custom tools, crawling, scheduled tasks, and Feishu notifications. This gives limited clues about permissions and data flows, but provides no concrete permission configuration, confirmation gates, credential protections, outbound controls, or recovery details, so each trust criterion receives 1. Infrastructure and tool sources are named, but per-tool attribution and security maintenance practices are not established.
The README describes a multi-service architecture, startup commands, and Docker Compose deployment, and provides commands for service status and logs. However, it inconsistently spells the release compose file extension (.yml versus .yaml), and does not explain dependency pinning, recovery when services are unavailable, or user-facing error handling, so each criterion receives 1.
Researchers, developers, and students are named as audiences, and the README lists disciplinary tools, skills, deployment options, and task types. It does not clearly define capability boundaries, triggering conditions, or environment-specific constraints; its claim of minimal setup also sits alongside Docker, RAM, and API resource requirements. Each criterion receives 1.
The README has a contents list, project structure, feature descriptions, installation steps, useful commands, uninstall instructions, and an MIT license link, supporting good information organization and license presentation. Installation notes include default administrator credentials and a password-change reminder, but offer limited detail on secrets, security configuration, or troubleshooting. Compose filenames are inconsistent; no FAQ, known limitations, or clear maintenance and update policy are provided. Version information is limited to desktop v0.0.4, so related scores remain modest.
The README describes multidisciplinary scientific tools, scheduled tasks, file management, and deliverables in PDF, DOCX, PPTX, and XLSX, suggesting useful research outputs and some added value. Claims such as 1,900+ tools, security, transparency, and privacy are primarily assertions in the README. It provides no cost, resource-use, or comparative outcome evidence, so marginal value and cost-benefit receive 1.
The README names components, directories, deployment commands, and features, making some product claims traceable to specific components. The supplied material contains only the README, with no code, tests, or other sources for corroboration, and promotional claims are not consistently separated from verifiable facts. Claim traceability and fact/inference separation receive 1; cross-source corroboration receives 0.
- The README gives the default administrator credentials admin/admin123 and only advises changing the password after first login; confirm how the default is disabled or replaced before deployment.
- Although it claims data stays local, the README also lists web search, crawling, external scientific tools, LLM APIs, a cloud version, and Feishu notifications; verify where requests, credentials, and results actually go.
- Code execution, shell access, file writing, and automatic tool creation/testing are listed, but confirmation controls, container permissions, and recovery behavior are not described.
- The README spells the release Compose filename inconsistently, and the supplied material does not corroborate its security, privacy, or tool-count claims.