Integration Proposal: CAJAL — Local Scientific Paper Model for ShellGPT

Integration Proposal: CAJAL — Local Scientific Paper Model for ShellGPT

Hi ShellGPT team! I'm Francisco (@Agnuxo1) from the P2PCLAW research ecosystem.

Proposal

Add CAJAL as a model option within ShellGPT for terminal-based scientific paper generation.

What is CAJAL?

• 4B parameter local model (GGUF), runs on consumer GPUs (24GB VRAM)
• Purpose-built for academic writing: LaTeX/Markdown, structured sections, citations
• Fully offline — no API dependency via llama.cpp/Ollama/vLLM
• Apache 2.0 open source
• HF: https://huggingface.co/Agnuxo/CAJAL-4B-P2PCLAW
• GitHub: https://github.com/Agnuxo1/CAJAL

Why ShellGPT + CAJAL?

ShellGPT's terminal interface + CAJAL = perfect for researchers who live in the terminal:

1. Command-line paper drafting: sgpt --model cajal "Write abstract on quantum computing"
2. Pipe workflows: cat experiment_results.txt | sgpt --model cajal "Generate Results section"
3. Shell integration: Generate papers directly from shell scripts and data pipelines
4. Local-first: All processing stays on the user's machine

Integration approach

• Model integration: Add CAJAL to the supported local models (Ollama-compatible)
• System prompt: Pre-configured academic writing system prompt for CAJAL
• Example workflows: Terminal-based paper generation examples

Questions

1. Would CAJAL fit as a model option or require custom system prompt configuration?
2. Is there interest in specialized academic/research use cases from the community?
3. Should I start with a documentation PR or direct model integration?

Thanks for the awesome terminal tool!

---

P2PCLAW: Decentralized research network (14 agents, post-quantum crypto, Lean 4).