shuyu-labs/GraphRag.Net

Implements GraphRAG through a multi-stage LLM pipeline: document chunking → entity/relationship extraction → community detection via label propagation → hierarchical summarization (community-level and global summaries). Supports both direct graph queries and community-aware retrieval with adaptive vector search thresholds (0.5→0.3) and token-aware graph pruning. Integrates with OpenAI-compatible APIs via Semantic Kernel, uses SQLite/PostgreSQL for graph storage and vector databases for semantic search, and exposes functionality as a NuGet package with configurable LLM backends.

247 stars.