Add CAJAL — 100% Offline Local LLM for Scientific Paper Writing

[Agnuxo1]

Proposal: Add CAJAL — 100% Offline Scientific Paper Generator

Project: CAJAL | HuggingFace | PaperClaw VS Code Extension
Category: Local / Offline AI — Scientific Writing

What is CAJAL?

CAJAL is a family of locally-runnable language models (4B–9B parameters) fine-tuned from Qwen3.6 on 500K+ academic papers for scientific writing:

• 100% offline — Runs entirely on consumer GPU (RTX 3090 24GB). No internet, no API keys, no cloud
• AI Tribunal peer review — Built-in multi-dimensional scoring (originality, methodology, clarity, reproducibility)
• PaperClaw VS Code extension — Turns VS Code into a scientific paper editor with structured sections, citations, and arXiv export
• Apache 2.0 — Fully open source, weights on HuggingFace

Why it fits this list

This list curates offline/local AI tools. CAJAL is the only local LLM specifically designed for scientific paper generation — not chat, not coding, but academic writing. Critical for researchers who:

• Work with sensitive data (can't send to cloud APIs)
• Need reproducible paper generation
• Want to own their research pipeline end-to-end

Benchmarks

• Ranked Add CAJAL: Local Scientific Paper Generation Model (Offline, 4B-9B, Ranked #3) #3 on internal P2PCLAW benchmarks (below only Claude Sonnet 4.6)
• Outperforms GPT-4o-mini and Llama-3.1-8B on scientific reasoning
• 2.7× token compression vs. standard formatting

Ecosystem

Part of P2PCLAW — decentralized research network with 14 agents, Lean 4 verification, IPFS persistence. Production paper: arXiv:2604.19792.

Happy to submit a PR!