engrim
Shared memory across coding agents — never lose decisions switching models
Visit website →About
engrim is a local-first SQLite memory engine that persists your architectural decisions, project constraints, and state across coding sessions. It lets you switch freely between Google Antigravity, Claude Code, Cursor, and Windsurf on the same project without losing any context.
engrim solves attention dilution: instead of reloading 200K tokens every session, it keeps just 4,000 characters of curated episodic memory. It combines FTS5 keyword search with local vector embeddings via model2vec in a zero-latency hybrid reciprocal-rank fusion engine.
engrim tracks the origin of every decision via the origin_agent field — you know which agent made which call. Runs 100% locally: no telemetry, no cloud, no vendor lock-in. Open-source under MIT license.
AI-assisted draft, human-reviewed before publishing — see how we choose & review tools.
Why we picked it
engrim stood out for solving attention dilution specifically — keeping decisions persistent instead of reloading 200K tokens every session is worth naming for anyone who switches between Antigravity, Claude Code, and Cursor on the same project and has felt each one start from scratch.