OmicsClaw
Conversational & memory-enabled AI research partner for multi-omics analysis. From biological idea to full research paper.
Install / Use
/learn @TianGzlab/OmicsClawREADME
OmicsClaw
AI research assistant that remembers. OmicsClaw transforms multi-omics analysis from repetitive command execution into natural conversations with a persistent partner that tracks your datasets, learns your methods, and resumes interrupted workflows across sessions.
<h3>⚡ Unified Control, Different Surfaces</h3> <table> <tr> <th width="75%"><p align="center">🖥️ CLI / TUI</p></th> <th width="25%"><p align="center">📱 Mobile (Feishu)</p></th> </tr> <tr> <td align="center"> <video src="https://github.com/user-attachments/assets/a24b16b8-dc72-439a-8fcd-d0c0623a4c8a" autoplay loop muted playsinline width="100%"> <a href="https://github.com/user-attachments/assets/a24b16b8-dc72-439a-8fcd-d0c0623a4c8a">View CLI demo</a> </video> </td> <td align="center"> <video src="https://github.com/user-attachments/assets/0ccb21f8-6aa9-45ec-b50d-44146566e64e" width="100%" autoplay loop muted playsinline> <a href="https://github.com/user-attachments/assets/0ccb21f8-6aa9-45ec-b50d-44146566e64e">View mobile demo</a> </video> </td> </tr> </table>[!NOTE] 🚀 v0.1.0 正式版发布 / Official v0.1.0 Release
经过充分的开发与严格测试，OmicsClaw v0.1.0 现已正式发布！在这一里程碑大版本中，我们提升了交互式自然语言分析的体验，并引入了直观的原生记忆管理面板（Memory Explorer），提供了覆盖 6 个组学领域的 72 个内置原生技能。欢迎下载体验，任何问题与建议请通过 GitHub Issues 提交。期待您的反馈！
OmicsClaw v0.1.0 is officially released! This milestone version completes the core architecture, elevating the interactive natural language analysis experience, introducing a native Memory Explorer dashboard, and providing robust execution of 72 built-in skills across 6 omics domains. Try it now and share your feedback via GitHub Issues.
Why OmicsClaw?
Traditional tools make you repeat yourself. Every session starts from zero: re-upload data, re-explain context, re-run preprocessing. OmicsClaw remembers.
✨ Features
- 🧠 Persistent Memory — Context, preferences, and analysis history survive across sessions.
- 🛠️ Extensibility (MCP & Skill Builder) — Natively integrates Model Context Protocol (MCP) servers and features
omics-skill-builderto automate custom analysis deployment. - 🌐 Multi-Provider — Anthropic, OpenAI, DeepSeek, or local LLMs — one config to switch.
- 📱 Multi-Channel — CLI as the hub; Telegram, Feishu, and more — one agent session.
- 🔄 Workflow Continuity — Resume interrupted analyses, track lineage, and avoid redundant computation.
- 🔒 Privacy-First — All processing is local; memory stores metadata only (no raw data uploads).
- 🎯 Smart Routing — Natural language routed to the appropriate analysis automatically.
- 🧬 Multi-Omics Coverage — 72 predefined skills across spatial, single-cell, genomics, proteomics, metabolomics, bulk RNA-seq, literature and orchestration.
What makes it different:
| Traditional Tools | OmicsClaw | |-------------------|-----------| | Re-upload data every session | Remembers file paths & metadata | | Forget analysis history | Tracks full lineage (preprocess → cluster → DE) | | Repeat parameters manually | Learns & applies your preferences | | CLI-only, steep learning curve | Chat interface + CLI | | Stateless execution | Persistent research partner |
📖 Deep dive: See docs/MEMORY_SYSTEM.md for detailed comparison of memory vs. stateless workflows.
📦 Installation
To prevent dependency conflicts, we strongly recommend installing OmicsClaw inside a virtual environment. You can use either the standard venv or the ultra-fast uv.
Option A: Using standard venv
# 1. Create a virtual environment
python3 -m venv .venv
# 2. Activate it
source .venv/bin/activate
Option B: Using uv (Ultrafast)
# 1. Install uv (if you don't have it)
curl -LsSf https://astral.sh/uv/install.sh | sh
# 2. Create and activate virtual environment
uv venv
source .venv/bin/activate
# Clone the repository
git clone https://github.com/TianGzlab/OmicsClaw.git
cd OmicsClaw
# Install core system operations
pip install -e .
# Optional: Install Interactive TUI & Bot capabilities
# Includes prompt-toolkit/Textual plus the LLM client stack used by interactive mode
pip install -e ".[tui]"
pip install -r bot/requirements.txt # If you want messaging channels
Advanced installation tiers:
pip install -e .— Core system operationspip install -e ".[<domain>]"— Where<domain>isspatial,singlecell,genomics,proteomics,metabolomics, orbulkrnapip install -e ".[spatial-domains]"— Standalone Deep Learning Layer forSpaGCNandSTAGATEpip install -e ".[full]"— All domain extras and optional method backends across all domains
Check your installation status anytime with python omicsclaw.py env.
🔑 Configuration
The Easiest Way (Interactive Setup): OmicsClaw provides a built-in interactive wizard that walks through LLM setup, shared runtime settings, graph memory options, and messaging channel credentials in one flow.
omicsclaw onboard # or use short alias: oc onboard
The wizard writes the project-root .env used by CLI, TUI, routing, and bot entrypoints.
OmicsClaw supports switching between multiple LLM engines with a single config change. It automatically loads the project-root .env file for CLI, TUI, routing, and bot entrypoints. If python-dotenv is not installed, it falls back to a built-in .env parser, so standard key/value configuration still works in lean installs.
For hosted providers, you can configure either:
LLM_API_KEY- a provider-specific key such as
DEEPSEEK_API_KEY,OPENAI_API_KEY, orANTHROPIC_API_KEY
1. DeepSeek (Default):
DEEPSEEK_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
2. Anthropic (Claude):
ANTHROPIC_API_KEY=sk-ant-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
# Automatically detects the key and defaults to claude-3-5-sonnet
3. OpenAI (GPT-4o):
OPENAI_API_KEY=sk-proj-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
4. Local LLM (Ollama): If you have strict data compliance requirements, you can run models entirely locally via Ollama. No API key is needed:
LLM_PROVIDER=ollama
OMICSCLAW_MODEL=qwen2.5:7b # Replace with your pulled model
LLM_BASE_URL=http://localhost:11434/v1
5. Custom OpenAI-compatible endpoint:
LLM_PROVIDER=custom
LLM_BASE_URL=https://your-endpoint.example.com/v1
OMICSCLAW_MODEL=your-model-name
LLM_API_KEY=sk-xxxxxxxxxxxxxxxx
</details>📖 Full Provider List: See
.env.examplefor instructions on configuring other engines like NVIDIA NIM, OpenRouter, DashScope, and custom endpoints.📖 Bot / channel config: See bot/README.md and bot/CHANNELS_SETUP.md for messaging channel credentials, allowlists, and runtime controls.
⚡ Quick Start
1. Chat Interface (Recommended)
# Start the Interactive Terminal Chat
omicsclaw interactive # or: omicsclaw chat
omicsclaw tui # or: oc tui
# OR start messaging channels as background frontends
python -m bot.run --channels feishu,telegram
📖 Bot Configuration Guide: See bot/README.md for detailed step-by-step instructions on configuring
.envand channel-specific credentials.
Chat with your data:
You: "Preprocess my Visium data"
Bot: ✅ [Runs QC, normalization, clustering]
💾 [Remembers: visium_sample.h5ad, 5000 spots, normalized]
[Next day]
You: "Find spatial domains"
Bot: 🧠 "Using your Visium data from yesterday (5000 spots, normalized).
Running domain detection..."
| Command | Description |
| ------- | ----------- |
| Analysis & Orchestration | |
| /run <skill> [...] | Run an analysis skill directly (e.g. /run spatial-domains --demo) |
| /skills [domain] | List all available analysis skills |
| /research | Launch multi-agent autonomous research pipeline |
| /install-skill | Add new custom skills or extension packs from local or GitHub |
| Workflow & Planning | |
| /plan | Interactively inspect or create the session's action plan |
| /tasks | View the structured execution steps for the current pipeline |
| /approve-plan | Approve the autonomous pipeline to proceed |
| /do-current-task | Proceed with the next execution step in the pipeline |
| Session & Context Memory | |
| /sessions | List all recent saved conversational workflows |
| /resume [id/tag] | Resume
