Hello, I am coding a cli tool in rust to make an optimized queryable code graph. Uses Treesitter, Rag for semantic search and coding cli integration via MCP Learn a lot by doing it, love it 😄 People start using it at work to reduce token consumption and for result speed https://jrollin.github.io/cartog/

@Alex Galea i have an apple m4 pro 24go i use ONNX with Qdrant/bge-small-en-v1.5-onnx-Q (quantized BGE-small, 384-dim, ~70MB) for semantic search For indexing a large Javascript project of 2K files (~400K LOC): - without RAG and LSP : 1,5s and 160MB RAM - with RAG embed : 17,5s and 1.6GB RAM For updating the indexes, i implemented merkle tree to invalidate parts of graphs with hash Nb: you can use ollama too with config file to use another model and your GPU

Claude code, used opencode with my claude subscription, but not possible anymore.. Like the way they iterate on cli