litegraphdb/litegraph

litegraph attempts to carve out a niche by combining three hot but distinct storage paradigms (Graph, Vector, Relational) into a 'lite' footprint tailored for AI agents. While the inclusion of Anthropic's Model Context Protocol (MCP) shows current awareness, the project's quantitative signals are concerning: 106 stars over 5 years indicates a very slow adoption curve and a lack of community momentum compared to competitors like LanceDB (for embedded vector) or Kuzu (for embedded graph). The 'multi-model' approach is a difficult architectural feat; being 'lite' often means sacrificing the deep optimization found in specialized engines. The primary threat is twofold: 1) Frontier labs like Anthropic or OpenAI are increasingly building their own 'Memory' and 'Context' layers, potentially making third-party 'agent databases' redundant. 2) Established lightweight databases (like DuckDB or SQLite via extensions) are rapidly adding vector and graph capabilities, leveraging much larger ecosystems. Without a significant breakthrough in performance or a unique 'killer feature' beyond just being a wrapper for multiple storage types, it remains a high-risk project for displacement by more specialized or better-funded infrastructure.