Summary

NanoResearch is an end-to-end autonomous AI research engine that automates the entire scientific workflow. It distinguishes itself by executing actual computational experiments on GPU clusters, generating code, analyzing results, creating figures, and writing LaTeX papers grounded in real data, enabling researchers to focus on innovation rather than manual execution.

How It Works

The system orchestrates a nine-stage pipeline: IDEATION, PLANNING, SETUP, CODING, EXECUTION (local/SLURM GPU training), ANALYSIS, FIGURE_GEN, WRITING, and REVIEW. Its core innovation is performing real computational experiments, ensuring all paper components are derived from actual results. The pipeline supports breakpoint resumption and flexible model routing for different LLMs per stage.

Quick Start & Requirements

• Installation: Clone the repository and install with pip install -e ".[dev]".
• Configuration: Requires ~/.nanoresearch/config.json with an OpenAI-compatible API endpoint and API key.
• Prerequisites: Python 3.10+, OpenAI-compatible API endpoint, tectonic (recommended) or pdflatex for PDF compilation. Optional: SLURM cluster, OpenAlex/Semantic Scholar API keys.
• Resources: Estimated costs range from ~$0.5-$1 (paper-only) to ~$10-$20 (full pipeline). Execution time is 2-5 hours for the full pipeline.
• Links: Quick Start, Claude Code Mode.

Highlighted Details

• Full Research Automation: Covers literature search, experiment design, code generation, GPU execution, analysis, figure generation, and LaTeX paper writing.
• Grounded in Real Experiments: All outputs are based on actual computational results, not LLM fabrications.
• Resumable Pipeline: Supports breakpoint resumption for recovery from failures.
• Hybrid Model Support: Enables configuring different LLMs for specific stages.
• Multiple Interfaces: Accessible via Python CLI, Claude Code (no API keys), and Feishu (Lark) bot.

Maintenance & Community

The project is licensed under the MIT License. Acknowledgements mention claude-scholar and nanobot. No direct community links or active contributor details are provided in the README.

Licensing & Compatibility

MIT License: Permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

Generated papers are high-quality drafts requiring human review. Relies on external API endpoints for LLMs, necessitating configuration and incurring costs. PDF compilation may require specific TeX installations (tectonic recommended).