Add CAJAL: Local Scientific Paper Generation Agent

[Agnuxo1]

Proposal: Add CAJAL as a Scientific Data Agent

CAJAL is a local LLM agent specialized in scientific paper generation (4B/9B parameters) that bridges data analysis and academic writing.

What it does

• Data → Paper Pipeline: Takes research data, literature context, or experimental results and generates structured academic papers
• Section-Aware Generation: Uses special tokens [ABSTRACT], [INTRO], [METHODS], [RESULTS], [DISCUSSION], [CONCLUSIONS], [REFERENCES] for structured output
• Peer Review Simulation: Scores generated papers (0-10) on Originality, Methodology, Clarity, Impact
• Citation Validation: Verifies all references against arXiv API
• 100% Local: Runs via Ollama/llama.cpp — zero API cost, works offline

Why it fits this list

This list covers Proto-L3 data agents. CAJAL is a specialized agent for the scientific research domain:

• It orchestrates a full pipeline from research context → structured paper
• It validates its own output (citation check + peer review scoring)
• It's open-source and runs locally (8GB VRAM for 4B, 16GB for 9B)

Links

• GitHub: https://github.com/Agnuxo1/CAJAL
• HuggingFace: https://huggingface.co/Agnuxo/CAJAL-9B-P2PCLAW
• Paper: https://arxiv.org/pdf/2604.19792
• PyPI: https://pypi.org/project/cajal-p2pclaw/

Quick Start

huggingface-cli download Agnuxo/CAJAL-9B-P2PCLAW --local-dir ./cajal
ollama create cajal -f ./cajal/Modelfile
ollama run cajal

MIT licensed.

[derrickzhuyz]

Thanks for sharing CAJAL with us!

This looks like an interesting and well-executed project, especially for scientific writing and paper generation. After reviewing it, we feel that CAJAL is more aligned with a scientific writing assistant than with the core scope of this list, which focuses on data-centric challenges.

For this reason, we may not include it in the list for now. Thanks again for the suggestion!