CtxOS Cookbook Engine

Convert Declarative AI Recipes → Executable Infrastructure

🧠 Philosophy

This repo is an AI DevOps compiler that transforms YAML recipes into executable Jupyter notebooks, Docker configurations, and deployment manifests. It is NOT just a notebook generator.

🏗 Structure

CtxOS/cookbook/
│
├── cookbooks/                # Declarative recipes
│   ├── llm_finetune.yaml
│   ├── rag_pipeline.yaml
│   └── agent_builder.yaml
│
├── ctxcook/                  # Core engine (Python package)
│   ├── __init__.py
│   ├── parser.py
│   ├── validator.py
│   ├── generator.py
│   ├── notebook.py
│   ├── exporter.py
│   └── cli.py
│
├── templates/
│   ├── colab_template.ipynb
│   ├── docker_template.j2
│   └── hf_deploy_template.j2
│
├── tests/
│   ├── test_parser.py
│   └── test_validator.py
│
├── pyproject.toml
├── README.md
└── Makefile

🚀 Quick Start

Installation

# Install locally
pip install -e .

This makes the ctxcook CLI available globally.

Usage

# Build a notebook from a recipe
ctxcook build cookbooks/llm_finetune.yaml --output notebooks/finetune.ipynb

# Help
ctxcook --help
ctxcook build --help

📖 Example Recipes

LLM Fine-tuning Recipe

name: llm_finetune_basic

model:
  base: mistralai/Mistral-7B-v0.1
  quantization: 4bit

dataset:
  source: huggingface
  name: secpatch/train

RAG Pipeline Recipe

name: rag_pipeline_basic

model:
  base: sentence-transformers/all-MiniLM-L6-v2
  type: embedding

dataset:
  source: local
  path: ./documents

retrieval:
  chunk_size: 512
  overlap: 50

Agent Builder Recipe

name: agent_builder_basic

model:
  base: gpt-3.5-turbo
  provider: openai

tools:
  - name: calculator
    type: function
  - name: web_search
    type: api

agent:
  type: conversational
  memory: true

🔧 Development

Running Tests

python -m pytest tests/

Code Style

black ctxcook/
flake8 ctxcook/

🚀 Production Upgrade Plan

Phase 1

• Add Jinja2 templating
• Add environment config (Colab / Docker / Local)
• Add version pinning

Phase 2

• DAG builder (convert recipe → pipeline graph)
• Add cost estimation
• Add GPU selection logic

Phase 3

• Integrate with CtxAI runtime
• Allow prompt → auto-generate cookbook YAML
• Add marketplace registry support

📄 License

MIT License - see LICENSE file for details.