Overview
What this is
The compact read before the technical details.
A local-first semantic file search engine combining BM25 keyword search with FAISS-powered vector search for instant document retrieval. Built with FastAPI and ONNX Runtime for fast embeddings, it indexes your local files and lets you search by meaning, not just keywords. Supports multiple file types, runs entirely offline, and returns ranked results in milliseconds.
Capabilities
What it actually does
The useful parts, pulled out of the paragraph wall.
Hybrid search engine fusing BM25 keyword matching with FAISS-powered dense vector embeddings using reciprocal rank fusion — gets you both exact-match precision and semantic recall
Local-first architecture: all indexing, embedding, and search runs on-device via ONNX Runtime with no cloud dependency
Hardware-adaptive system that auto-detects capabilities and throttles between eco, balanced, and performance modes
Multi-format pipeline handling PDFs, DOCX, PPTX, XLSX, and code files with syntax-aware chunking via Docling and Chonkie
Implementation
Technology with jobs attached
Names are less useful than responsibilities. This is what each piece is doing.
FastAPI
Async backend API serving search, indexing, and stats endpoints
Sentence Transformers
Generates dense vector embeddings for semantic search queries and documents
FAISS
High-performance approximate nearest neighbor search over document embeddings
Elasticsearch
BM25 keyword retrieval engine for traditional term-based matching
ONNX Runtime
Local inference runtime for transformer models without GPU requirements
Docling
Document parsing and content extraction for PDFs, PPTX, and XLSX files
Chonkie
Text chunking library with syntax-aware splitting for code files
React 19
Frontend UI with component-based search interface