agi-memory: SQLite-only persistent memory MCP server for coding assistants, 32MB RAM
Rude_Gate7599 · reddit · 2026-09-14
- A developer released agi-memory, an MCP server giving persistent cross-session memory to coding assistants like Claude Code, Cursor, Codex, Windsurf, Aider and Cline — it saves decisions, bug fixes and codebase notes to a local file and feeds them back into whichever assistant opens next.
- Radical minimalism: Python stdlib + SQLite only, no vector DB, no embeddings, no daemon. 32MB RAM, sub-millisecond lookups, fully offline — versus comparable tools that ship 500MB of ML libraries and take 200–500ms per lookup.
- The author is upfront about the tradeoff: keyword search can't bridge synonyms like embeddings can (searching "login" won't find a note saying "authentication"), and ships a public eval suite measuring recall on rephrased queries, including categories where it performs badly.
- The project is one week old and the author is actively soliciting feedback on where it breaks.
More from coding & agent
- Agent Scorecard: Five Numbers to Tell Whether Your AI Agents Actually Work — kashifmanzoor · 2026-09-14
- TraceHub-MCP: One MCP Toolset to Query Traces Across Jaeger, Datadog, Sentry and More — Ok_Bell443 · 2026-09-14
- OpenViking launches as an open-source context database for AI agents — adnan_hashmi · 2026-09-14
- Git worktrees ~80% faster via APFS/btrfs/XFS/ReFS folder clones, with big disk savings — steipete · 2026-09-14
- Dev burns ~80% of a 20x AI sub debugging cascading AWS failures — generativist · 2026-09-14
- Microsoft's AgentRx paper pinpoints where long agent traces really fail — JohnAlexander · 2026-09-14