Show HN: Engrim – A universal, local-first SQLite memory engine for AI CLIs
summary
Engrim is a local-first SQLite memory engine designed to prevent context loss and attention dilution when switching between different AI coding agents and models. By decoupling project intelligence from specific AI vendors, it allows developers to maintain architectural decisions and project state across tools like Claude Code, Cursor, and Google Antigravity. The system uses a hybrid search engine to provide curated, low-token memory packs that significantly reduce the cost and noise of reloading large context windows.
key points
- The tool implements a hybrid retrieval system combining SQLite FTS5 for keyword search and model2vec for static vector embeddings to load relevant context with zero latency.
- It features agent provenance tracking, which records which specific AI agent made a decision, ensuring a clear history of architectural changes across a multi-agent workflow.
- Engrim integrates via the Model Context Protocol (MCP) and lifecycle hooks, enabling agents to automatically save and recall decisions without manual developer intervention.
who should read it
Software developers who use multiple AI coding assistants and want to maintain a consistent project state without repeatedly re-explaining constraints to different models.
discussion positive
consensus
Commenters generally appreciate the local-first, SQLite-based approach to agent memory, valuing the privacy, speed, and lack of cloud lock-in. There is a shared understanding that fresh sessions are optimal for avoiding context degradation, provided that key architectural decisions are preserved. However, some users remain skeptical about whether autonomous agent memory management actually works in practice or if it simply results in noise.
top insight
The distinction between permanent project documentation in Git and a 'desk scratchpad' for session context is crucial. While architectural records belong in version control for PR reviews, a tool like engrim serves as a volatile bridge that allows developers to clear sessions frequently without losing the immediate momentum or micro-decisions of a task.
Generated by an LLM from the article and thread. Check the sources before relying on it.