Smriti
A local memory layer for AI agents that stores everything in a single SQLite file, tracks when facts change without deleting old versions, and exposes six typed tools through a one-command MCP server.
Smriti gives AI agents persistent memory backed by one SQLite file, with no Postgres, Neo4j, Docker or cloud account required; the core runs on the Python standard library plus numpy. Its read engine keeps whole conversation turns rather than fragments, resolves relative dates, and groups evidence by session so the model receives complete, dated context instead of disconnected snippets.
Facts are handled with bi-temporal supersession: a changed fact replaces the old one in the model's answers without deleting the original record, and validity windows let a query like 'where do I live now' and 'where did I live in March' both resolve correctly from the same store. A one-command MCP server exposes six typed memory tools to Claude Code, Cursor and other MCP clients.
An optional enterprise package adds audit receipts, data retention, legal holds and verified export packs on top of the open-source core, which is Apache-2.0 licensed.