0 Prerequisites
- ☑ Python 3.10+
- ☑ Docker with NVIDIA Container Toolkit (for GPU experiments)
- ☑ An OpenAI-compatible API key (Azure OpenAI, OpenAI, or local LLM)
- ☑ NVIDIA GPU with 8GB+ VRAM (optional, for Docker sandbox)
1 Clone the Repository
git clone https://github.com/aiming-lab/AutoResearchClaw.git
cd AutoResearchClaw
2 Install Dependencies
pip install -e .
This installs the researchclaw package and all required dependencies.
3 Configure Your LLM
Create a YAML config file (e.g., config.yaml) with your LLM settings:
project:
name: "my-first-paper"
mode: "docs-first"
research:
topic: "Your research topic here"
llm:
provider: "openai-compatible"
base_url: "https://api.openai.com/v1"
api_key_env: "OPENAI_API_KEY"
experiment:
backend: "docker"
timeout_sec: 1800
4 Set Your API Key
export OPENAI_API_KEY="sk-your-key-here"
5 Build the Docker Image (Optional)
If using the Docker sandbox backend for GPU-accelerated experiments:
docker build -t researchclaw-sandbox -f researchclaw/docker/Dockerfile .
6 Run Your First Paper
python -m researchclaw run --config config.yaml
The pipeline will execute all 23 stages autonomously. Output will be saved
to the output/ directory including the paper PDF, LaTeX source,
experiment code, and charts.
7 Review Your Paper
After the pipeline completes, find your generated paper at:
output/<run-id>/
paper.pdf
paper.tex
references.bib
code/main.py
charts/
results.json
Tips
- Use GPT-4.1 or newer for best paper quality
- Set
timeout_sec: 3600 for complex experiments
- For Azure OpenAI, set
provider: "azure_openai" and configure your endpoint
- The pipeline caches literature results, so re-runs with the same topic are faster
- Run
python -m pytest tests/ -v to verify your installation