microsoft/graphrag

GraphRAG is a mature, well-adopted Microsoft project (32k stars, 3.3k forks) that represents a significant shift in RAG architecture from simple semantic search to structured knowledge graphs. The modular design, active maintenance, and integration with major LLM platforms establish it as infrastructure-grade for enterprise RAG pipelines. DEFENSIBILITY: Scores 8 due to: (1) Network effects and ecosystem lock-in—adopted across enterprises and embedded in multiple downstream projects; (2) Deep technical moat in the graph extraction + LLM retrieval pipeline that requires domain expertise to replicate; (3) Community momentum with 3.3k forks indicating active usage as a foundation; (4) Negative velocity (-2/hr suggests stabilization post-launch, not abandonment). The remaining risk is that core concepts are not patented and competitors can implement similar approaches. PLATFORM DOMINATION RISK: High. Microsoft owns this project and Azure/OpenAI integration is native. Google (Vertex AI + BigQuery graph capabilities), AWS (Bedrock + Neptune), and OpenAI (Assistants API with knowledge bases) are all building competing RAG layers. Anthropic's function-calling and prompt caching reduce the need for explicit graph retrieval. The threat is not displacement of the code, but commoditization of graph RAG as a standard LLM feature within 1-2 years. MARKET CONSOLIDATION RISK: Medium. The RAG market is consolidating around vector databases (Pinecone, Weaviate, Qdrant) and prompt orchestration (LangChain, LlamaIndex). GraphRAG differentiation is strong now but no single incumbent owns 'graph RAG'—it's an architectural pattern, not a proprietary technology. Acquisition risk is low (already owned by Microsoft); displacement risk is moderate as vector DB vendors and LLM platforms add graph capabilities. DISPLACEMENT HORIZON: 1-2 years. Graph RAG is actively being commoditized by major platforms. Within 2 years, most enterprise LLM applications will have native graph retrieval via their chosen platform. GraphRAG's staying power depends on remaining a go-to open-source reference implementation and tool for users who want fine-grained control over graph extraction and retrieval logic. NOVELTY: Novel combination. Graph-based retrieval is not new (knowledge graphs, semantic networks exist), nor is RAG new (Retrieval-Augmented Generation is well-established). The contribution is integrating LLM-driven entity/relationship extraction with graph-based retrieval to improve context quality—this is a meaningful innovation in the RAG space, not a breakthrough.