A high-speed, local-model codebase indexer with a live-updating knowledge graph, semantic/regex/fuzzy search, code intelligence, and a grounded "ask the codebase" assistant — powered by Qwen3-Embedding-0.6B and LFM2.5-1.2B-Instruct running locally via embedded llama.cpp.
The project lives in graphindex/. See
graphindex/README.md for full docs.
cd graphindex
python -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt && pip install -e .
graphindex serve . # live WebUI + API at http://localhost:8000$env:GRAPHINDEX_BACKEND="llamacpp" ; $env:LLAMA_CUDA=0 ; $env:GRAPHINDEX_LOG_LEVEL="INFO" ; $env:GRAPHINDEX_LLAMA_LOG="warn" ; python .\graphindex\graphindex\cli.py serve graphindex/ --watch | tee log.txt