Memory Vault
A self-hosted memory database for AI applications, combining Postgres and pgvector with hybrid search and a knowledge graph so decisions and context persist across sessions with an AI assistant.
Memory Vault is a persistent memory layer for AI assistants, storing decisions, conversations, notes and project context in a single self-hosted Postgres database that combines semantic vector search with keyword search. An assistant can recall and store memories during a session through MCP, a local large language model can be used to chat with the stored memories directly, or a developer can build a separate application on top of its REST API.
It runs as a Docker Compose stack that includes Postgres with the pgvector extension and runs its own migrations on first start, or it can be run without Docker using Python and a locally installed pgvector-enabled Postgres. As of its first stable release, its REST API endpoints, MCP tool signatures and database schema are treated as a stable public surface, with breaking changes reserved for major version increases.