Facthouse, a local memory engine for AI tools

Field notes

Facthouse separates conversational memory into session events, extracted facts, and integrated knowledge, with entities linked across people, organisations, projects, places, and products. Its retrieval layer combines BM25, structured domain matching, and entity-graph paths through Reciprocal Rank Fusion, while embeddings remain optional and off by default. The documented setup runs locally through Node and SQLite, and the site provides separate paths for copying Claude Code or Cursor transcripts versus recording facts from other MCP clients.