# PROJECT_CONTEXT.md

## Project Overview
LocalFoodAI is a local food AI that provides complete nutritional information on foods and can generate menu proposals based on user specifications. It runs entirely on a local Ubuntu 24.04 VM (8 vCPU, 30 GB RAM, no GPU). No user data leaves the server. The backend is Python-based.

## Tech Stack
- **Operating System:** Ubuntu 24.04 (VM)
- **Backend:** Python 3.11+
- **Database:** SQLite (local, no cloud)
- **Local LLM:** Llama 3.1 8B (quantized via Ollama, Q4_K_M or equivalent)
  - CPU-only compatible
  - Fits in 30 GB RAM with quantization
  - Instruction-following tuned
  - Open-source license (compatible with student projects)
- **Local Web Search Tool:** SearXNG (fully local, anonymous)
- **Version Control:** Git via Gogs on git.btshub.lu
- **CI / Deployment:** Antigravity Agent Manager handles task execution
- **LLM Hosting:** Ollama local instance, no cloud APIs

## Rules & Constraints
- **No external APIs or cloud services** for computation or data fetching
- **All data and computation must remain on the local VM**
- **All commits must be traceable to a Taiga Task ID**
- Antigravity must **read this file before starting any task** to avoid hallucinating cloud-based solutions
- Model and backend selection must fit **VM constraints** (CPU-only, RAM limit)

## Best Practices
- Use quantized models for CPU efficiency
- Verify all AI-generated Python or database logic before approving commits
- Test database queries and prompt logic locally before integrating
- Attach all artifacts (Implementation Plans, task lists, browser recordings) to the corresponding Taiga task
- Always include the TG-<ID> prefix in commit messages