OpenLAIR/dr-claw

Dr. Claw – An Open‑Source AI Research Assistant

What it is – Dr. Claw is a full‑stack, model‑agnostic workspace that helps researchers move a project from the initial idea all the way to a publishable paper (and even the accompanying slides). It bundles a web UI, a CLI, and a desktop‑app front‑end around a set of “research skills” that are executed by LLM agents such as Claude Code, Gemini CLI, Codex, or any model reachable via OpenRouter.

Key capabilities

Feature	What you get
Research Lab dashboard	Structured view of the whole pipeline (Survey → Ideation → Experiment → Publication → Promotion) with automatic artifact folders (literature reviews, code, analysis, LaTeX draft, slide deck).
Auto‑Research Hub	One‑click “run everything” – the agents generate a task list, execute it sequentially, and email you when the run finishes.
100+ built‑in skills	Prompt‑templates for literature survey, idea generation, experiment implementation, result analysis, paper writing, rebuttal drafting, etc. Agents discover and invoke the appropriate skill automatically.
Multi‑model backend	Switch between Claude Code, Gemini CLI, Codex, or any OpenRouter model (GPT‑5, Llama, DeepSeek, Kimi, …) with a single API key.
Chat‑driven workflow	Describe your research goal in the chat; the inno-pipeline-planner skill turns it into a concrete brief and task list without manual templates.
Integrated file & Git explorer	Browse, edit, stage, commit, and switch branches from inside the UI; a trash‑bin protects accidental deletions.
News dashboard	Unified, scored feed pulling from arXiv, Hugging Face, GitHub trending, WeChat public accounts, X, and Xiaohongshu.
Terminal‑only mode	dr-claw chat gives a lightweight, tool‑calling session in any shell, useful on headless servers.
Desktop app (Electron)	Native macOS/Windows installers (.dmg, .exe) or npx dr-claw for zero‑setup launch.
OpenClaw mobile/voice integration	A JSON‑based control plane lets a companion app (OpenClaw) act as a voice‑enabled research secretary.

How to get started

1. Prerequisites – Node 20+ (v22 LTS recommended) and at least one of the supported agent CLIs (Claude Code, Gemini CLI, Codex). Some platforms need native build tools for node‑pty / better‑sqlite3.
2. Quick launch – Run npx dr-claw (or install globally with npm i -g dr-claw), then open http://localhost:3001 (or :5173 for dev mode) to create an account.
3. Desktop alternative – Download the beta installer from the Releases page, or run npm run desktop:dev for a hot‑reload Electron app.
4. CLI‑only – After the server is running, install the Python harness (pip install -e ./agent-harness) and log in with drclaw auth login. Use dr-claw chat or launch a specific agent (claude, gemini, codex) from a second terminal.
5. OpenRouter usage – Export OPENROUTER_API_KEY and pick a model with --model <slug> (e.g., anthropic/claude-sonnet-4). The same key works for the UI.

Installation snippets

# Zero‑setup (recommended)
npx dr-claw

# Or install globally
npm i -g dr-claw && dr-claw

# From source (development)
git clone https://github.com/OpenLAIR/dr-claw.git
cd dr-claw
npm install
cp .env.example .env   # edit as needed
npm run dev   # starts backend + Vite dev server

License – Dual‑licensed under GPL‑3.0 + AGPL‑3.0, meaning the core is free to use and modify, but any network‑deployed service must also share source code under the same terms.

Community & support – Slack, WeChat, and X links are provided in the badge bar; a FAQ lives under docs/faq.md. The project also ships a Claude Code plugin (dr-claw-plugin-cc).

Why it matters – While many tools focus on a single stage (e.g., literature search or code generation), Dr. Claw attempts to cover the entire research lifecycle with a single, extensible UI and a plug‑in‑friendly skill library. Its model‑agnostic design lets teams keep data and compute on‑premise, which is useful for privacy‑sensitive or resource‑intensive projects.

---

TL;DR – Dr. Claw is an open‑source, locally‑runnable research workspace that orchestrates LLM agents across the whole paper‑writing pipeline. Install with npx dr-claw or grab the desktop binary, configure one of the supported model back‑ends, and start chatting your way from idea to finished manuscript.