dockg

What problem does it solve?

DocKG eliminates the mismatch between fast text search and the need for structurally grounded, source-traceable answers across a document corpus.

Core Features & Use Cases

• Deterministic hybrid knowledge graph: Indexes documents into a hybrid semantic + structural graph (SQLite + LanceDB) for traceable retrieval instead of embedding-only RAG.
• Source-grounded passage packs for agents: Uses graph traversal (topics, entities, references, containment, sequencing) to generate packed excerpts that LLM agents can trust.
• Multipass corpus analysis pipeline: Runs a multi-phase pipeline for sampling, chunking, classification, embedding, and manifold analysis, with structured provenance outputs.
• MCP tool exposure: Publishes graph tools (e.g., graph_stats, query_docs, pack_docs, get_node) for Claude Code/Kilo Code/Copilot workflows and debugging.

Quick Start

Use this when building an index from a corpus folder by running: dockg build docs