BioClaw brings the power of computational biology directly into WhatsApp group chats. Researchers can run BLAST searches, render protein structures, generate publication-quality plots, perform sequencing QC, and search the literature — all through natural language messages.
Built on the NanoClaw architecture with bioinformatics tools and skills from the STELLA project, powered by the Claude Agent SDK.
Welcome to join our WeChat group to discuss and exchange ideas! Scan the QR code below to join:
Scan to join the BioClaw community
- Overview
- Quick Start
- Messaging channels
- Demo Examples
- System Architecture
- Included Tools
- Project Structure
- Citation
- License
The rapid growth of biomedical data, tools, and literature has created a fragmented research landscape that outpaces human expertise. Researchers frequently need to switch between command-line bioinformatics tools, visualization software, databases, and literature search engines — often across different machines and environments.
BioClaw addresses this by providing a conversational interface to a comprehensive bioinformatics toolkit. By messaging @Bioclaw in a WhatsApp group, researchers can:
- Sequence Analysis — Run BLAST searches against NCBI databases, align reads with BWA/minimap2, and call variants
- Quality Control — Generate FastQC reports on sequencing data with automated interpretation
- Structural Biology — Fetch and render 3D protein structures from PDB with PyMOL
- Data Visualization — Create volcano plots, heatmaps, and expression figures from CSV data
- Literature Search — Query PubMed for recent papers with structured summaries
- Workspace Management — Triage files, recommend analysis steps, and manage shared group workspaces
Results — including images, plots, and structured reports — are delivered directly back to the chat.
- macOS or Linux
- Node.js 20+
- Docker Desktop
- Anthropic API key or OpenRouter API key
# Clone the repository
git clone https://github.com/Runchuan-BU/BioClaw.git
cd BioClaw
# Install dependencies
npm install
# Configure environment (edit with your API keys — see model section below)
cp .env.example .env
# First time only: build the agent Docker image
docker build -t bioclaw-agent:latest container/
# Start BioClaw (WhatsApp: scan the QR code printed in the terminal on first run)
npm startBioClaw now supports two provider paths:
- Anthropic — default, keeps the original Claude Agent SDK flow
- OpenRouter / OpenAI-compatible — optional path for OpenRouter and similar
/chat/completionsproviders
Create a .env file in the project root and choose one of the following setups.
Option A — Anthropic (default)
ANTHROPIC_API_KEY=your_anthropic_keyOption B — OpenRouter (Gemini, DeepSeek, Claude, GPT, and more)
MODEL_PROVIDER=openrouter
OPENROUTER_API_KEY=sk-or-v1-your-key
OPENROUTER_BASE_URL=https://openrouter.ai/api/v1
OPENROUTER_MODEL=deepseek/deepseek-chat-v3.1Popular model IDs: deepseek/deepseek-chat-v3.1, google/gemini-2.5-flash, anthropic/claude-3.5-sonnet. Full list: openrouter.ai/models
Note: Use models that support tool calling (e.g. DeepSeek, Gemini, Claude). Session history is preserved within a container session; after idle timeout, a new container starts with a fresh context.
Generic OpenAI-compatible setup
MODEL_PROVIDER=openai-compatible
OPENAI_COMPATIBLE_API_KEY=your_api_key
OPENAI_COMPATIBLE_BASE_URL=https://your-provider.example/v1
OPENAI_COMPATIBLE_MODEL=your-model-nameAfter updating .env, restart BioClaw:
npm run devWhen a container starts, docker logs <container-name> will show which provider path is active.
In any connected chat, simply message:
@Bioclaw <your request>
Supported platforms include WhatsApp (default), WeCom, Discord, Slack (Socket Mode), and optional local web (browser) chat. Full setup steps, env vars, and disabling channels are in docs/CHANNELS.md (简体中文：docs/CHANNELS.zh-CN.md).
Optional Lab trace dashboard (SSE timeline, workspace tree): set ENABLE_DASHBOARD=true. With local web enabled it is served on the same port at /dashboard; otherwise it uses DASHBOARD_PORT on its own. See docs/DASHBOARD.md.
Just send the message to OpenClaw:
install https://github.com/Runchuan-BU/BioClaw
See the ExampleTask document for 6 ready-to-use demo prompts with expected outputs.
Below are live demonstrations of BioClaw handling real bioinformatics tasks via WhatsApp.
Analyze files in a shared workspace and recommend the best next analysis steps.
Run FastQC on paired-end FASTQ files and deliver the QC report with key findings.
BLAST a protein sequence against the NCBI nr database and return structured top hits.
Create a differential expression volcano plot from a CSV file and interpret the results.
Fetch a PDB structure, render it in rainbow coloring with PyMOL, and send the image.
Search PubMed for recent high-impact papers and provide structured summaries.
Visualize hydrogen bonds between a ligand and protein in PDB 1M17.
Show residues within 5Å of ligand AQ4 in PDB 1M17.
BioClaw is built on the NanoClaw container-based agent architecture, extended with biomedical tools and domain knowledge from the STELLA framework.
WhatsApp ──► Node.js Orchestrator ──► SQLite (state) ──► Docker Container
│
Claude Agent SDK
│
┌──────────┴──────────┐
│ Bioinformatics │
│ Toolbox │
├─────────────────────┤
│ BLAST+ │ SAMtools │
│ BWA │ BEDTools │
│ FastQC │ PyMOL │
│ minimap2│ seqtk │
├─────────────────────┤
│ Python Libraries │
├─────────────────────┤
│ BioPython │ pandas │
│ RDKit │ scanpy │
│ PyDESeq2 │ pysam │
│ matplotlib│ seaborn │
└─────────────────────┘
Key design principles (inherited from NanoClaw):
| Component | Description |
|---|---|
| Container Isolation | Each conversation group runs in its own Docker container with pre-installed bioinformatics tools |
| Filesystem IPC | Text and image results are communicated between the agent and orchestrator via the filesystem |
| Per-Group State | SQLite database tracks messages, sessions, and group-specific workspaces |
| Channel Agnostic | Channels self-register at startup; the orchestrator connects whichever ones have credentials |
Biomedical capabilities (attributed to STELLA):
The bioinformatics tool suite and domain-specific skills — including sequence analysis, structural biology, literature mining, and data visualization — draw from the tool ecosystem developed in the STELLA project, a self-evolving multi-agent framework for biomedical research.
| Tool | Purpose |
|---|---|
| BLAST+ | Sequence similarity search against NCBI databases |
| SAMtools | Manipulate alignments in SAM/BAM format |
| BEDTools | Genome arithmetic and interval manipulation |
| BWA | Burrows-Wheeler short read aligner |
| minimap2 | Long read and assembly alignment |
| FastQC | Sequencing quality control reports |
| fastp | FASTQ filtering and trimming (QC/preprocessing) |
| MultiQC | Aggregate QC reports into one summary |
| seqtk | FASTA/FASTQ file manipulation |
| seqkit | FASTA/FASTQ toolkit (extended) |
| BCFtools | Variant calling and VCF/BCF manipulation |
| tabix | Index/query compressed VCF/BED (bgzip/tabix) |
| pigz | Parallel gzip compression/decompression |
| SRA Toolkit | Download data from NCBI SRA (prefetch/fasterq-dump) |
| Salmon | RNA-seq transcript quantification |
| kallisto | RNA-seq transcript quantification |
| PyMOL | Molecular visualization and rendering |
| Library | Purpose |
|---|---|
| BioPython | Biological computation (sequences, PDB, BLAST parsing) |
| pandas / NumPy / SciPy | Data manipulation and scientific computing |
| matplotlib / seaborn | Publication-quality plotting |
| scikit-learn | Machine learning for biological data |
| RDKit | Cheminformatics and molecular descriptors |
| PyDESeq2 | Differential expression analysis |
| scanpy | Single-cell RNA-seq analysis |
| pysam | SAM/BAM file access from Python |
BioClaw/
├── src/ # Node orchestrator
├── container/ # Agent image + runner
├── groups/ # Per-group workspace & CLAUDE.md
├── docs/
│ ├── CHANNELS.md # Messaging platform setup (EN)
│ ├── CHANNELS.zh-CN.md # Messaging platform setup (ZH)
│ └── images/ # Doc screenshots
├── ExampleTask/ # Demo prompts + screenshots
├── bioclaw_logo.jpg # Project logo
└── README.md
BioClaw builds upon the STELLA framework. If you use BioClaw in your research, please cite:
@article{jin2025stella,
title={STELLA: Towards a Biomedical World Model with Self-Evolving Multimodal Agents},
author={Jin, Ruofan and Xu, Mingyang and Meng, Fei and Wan, Guancheng and Cai, Qingran and Jiang, Yize and Han, Jin and Chen, Yuanyuan and Lu, Wanqing and Wang, Mengyang and Lan, Zhiqian and Jiang, Yuxuan and Liu, Junhong and Wang, Dongyao and Cong, Le and Zhang, Zaixi},
journal={bioRxiv},
year={2025},
doi={10.1101/2025.07.01.662467}
}This project is licensed under the MIT License. See LICENSE for details.