FileSense
Local-first semantic file search — BM25 + FAISS hybrid, runs entirely offline.
PythonFAISSONNXSearch
Overview
Indexes local files and searches by meaning, not just keywords. BM25 handles exact matches; FAISS handles semantic recall; reciprocal rank fusion merges the two. Embeddings run on-device with ONNX Runtime. PDFs, Office docs, and code all get chunked appropriately.
Highlights
- Hybrid BM25 + dense retrieval with reciprocal rank fusion
- Fully local — indexing, embedding, and search stay on-device
- Eco / balanced / performance modes based on hardware
- PDF, DOCX, PPTX, XLSX, and code with syntax-aware chunking
Stack
- FastAPI
- Search and indexing API
- FAISS
- Vector nearest-neighbor search
- Elasticsearch
- BM25 keyword retrieval
- ONNX Runtime
- Local embedding inference
- React 19
- Search UI