16iro/airis

airis is a desktop local RAG engine and study/workspace: a Tauri app (React frontend + Rust backend) that ingests a book (PDF/Markdown/HTML/TXT), chunks and embeds it locally, and serves hybrid search results (sqlite-vec KNN + SQLite FTS5 + RRF) to an LLM subprocess (claude-code / gemini-cli / codex-cli) or via API keys. Embedding models (multilingual-e5-small, BGE-M3) run on-device (ONNX/fastembed); user data and tokens are not sent to external servers. Who uses it: privacy-conscious individuals or developers wanting an offline RAG tool for personal documents and books. It’s a single-author project currently in v0.4.x. Key features: - Parsers for MD/HTML/PDF/TXT with ICU sentence splitting and recursive chunking - Hybrid retrieval combining vector KNN + FTS5 + Reciprocal Rank Fusion - Local on-device embeddings (INT8/FP) with embedding_cache and response_cache - Robustness: WAL+transactions, resumable jobs, pause triggers (user/thermal/battery), resource limiting - LLM provider adapters, citation chips linking back to source sections - Buildable with Node/pnpm and Rust; models download on first indexing.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.