Omnichunk
Omnichunk splits source files and documents into deterministic, structure-aware chunks for RAG, embeddings, and LLM context windows.
Features
- Code, prose, and markup routing with tree-sitter where available
- Lossless byte and line ranges for reconstruction
- Optional multiformat loaders (
.ipynb, LaTeX, PDF/DOCX with extras) - Vector export helpers (Pinecone / Weaviate / Supabase shapes) without vendor SDKs
ChunkStorefor SQLite-backed incremental indexingomnichunk serve --mcpfor JSON-RPC tool access (stdlib HTTP server)