Purpose

Solves a problematic usage of roo code experimental codebase indexing functionality for llama.cpp. Idea is to have a simple api server for redirect of indexing request. These are problematic when used with llama.cpp as by default roo code extension tries to access /api/embed which is not supported endpoint in llama.cpp implementation.

Execution

before anythin start your llama.cpp server with embedding => e.g.

llama-server -m /root/hf_models/embeddings/magicunicorn/mxbai-embed-large-v1/mxbai-embed-large-v1-q8_0.gguf -c 512 --port 8081

Notice the port 8081, this is not default port of llama.cpp. I use it because I run this embedding as a separate instance to have it always available for roo code extension.

then run the api server

python api_server.py