aiming-lab/AutoResearchClaw

A 23-stage autonomous research pipeline that turns a single topic into a conference-ready paper — no human in the loop. You type a research idea; it runs literature discovery across OpenAlex, Semantic Scholar, and arXiv, generates hypotheses through multi-agent debate, executes hardware-aware sandbox experiments (NVIDIA CUDA, Apple MPS, or CPU auto-detected), performs statistical analysis, runs multi-agent peer review, and outputs publication-grade LaTeX targeting NeurIPS, ICML, and ICLR. When experiments fail, it self-heals. When hypotheses don't hold, it pivots. A 4-layer citation verification system eliminates hallucinated references. The MetaClaw integration means pipeline failures become structured lessons reused across future runs. Supports GPT-4o, DeepSeek, OpenRouter, and any OpenAI-compatible endpoint. Also works as an OpenClaw-compatible service you can launch from Claude Code, Copilot CLI, or Gemini CLI with a single message.