Mnemosyne
Local AI-agent memory over SQLite, with evidence history, corrections, dependency review and optional semantic retrieval through MCP, CLI, TypeScript or HTTP.
What happened lately
Mnemosyne timeline
Mnemosyne is the MIT-licensed TypeScript package mnemosy-ai, maintained in 28naem-del/mnemosyne. It is separate from the similarly named Python mnemosyne-memory project. The local engine stores records and their supporting evidence in SQLite and exposes MCP, CLI, SDK and HTTP interfaces.
The October 29 release candidate 3.0.0-rc.4 adds CLI review of dependency holds, hybrid-path retrieval bisect and encrypted backups that can include the consolidation signing key. npm’s stable channel remains 2.0.0. The tagged documentation requires Node.js 22.16 or later, describes author-tested Claude Code integration, and distinguishes documented Cursor and Hermes configurations from tested hosts. The local engine needs no model API key; optional retrieval models require explicit provisioning.
Workspace IDs are controller selectors rather than tenant authentication. Local database files remain plaintext even when encrypted backup bundles are used. AITrending has reviewed the documentation but has not reproduced the author’s demonstrations or host integrations.