Turn This Diet Agent Into Any Knowledge-Grounded Advisor

What it does

Healthy Diet AI Agent is a Bun/TypeScript backend that answers nutrition questions through chat and food-image analysis. It grounds responses in a local knowledge base plus uploaded documents using RAG, and can ingest new sources, extract knowledge graphs, and run background MOHW news sync. The whole thing ships as either a standalone Dockerized service with SQLite or as a plug-in module for existing Supabase stacks.

The interesting bit

The project began as internal glue for a specific diet-app ecosystem, but has since been untangled into an independent deployable agent. The most useful twist is that you can fork it and repurpose the core into another domain-specific advisor just by editing agent_config.json and swapping out the markdown knowledge files, without touching runtime code.

Key highlights

• Dual deployment modes: standalone SQLite with Docker and CLI, or Supabase integration for existing stacks
• RAG pipeline with document ingestion, reindexing, and knowledge graph extraction routes
• Agent behavior controlled by agent_config.json and markdown files like NUTRITION_RULES.md
• Background MOHW news sync
• Terminal CLI for direct local prompts without a frontend

Caveats

• Requires an external LLM endpoint via AI_API_URL or Google Gemini configuration; no built-in model
• Admin document-management routes require specific forwarded headers (X-Admin-User-Id and X-Admin-Role), not just a standard Authorization bearer

Verdict

Worth a look if you need a forkable template for domain-specific RAG agents with both HTTP and CLI interfaces. Skip it if you want a plug-and-play app with its own bundled model or a polished end-user frontend.