The engrim project introduces a universal, local-first SQLite memory engine that solves context dilution and vendor lock-in for AI development. This Zero cloud lock-in solution enables developers to switch seamlessly between Google Antigravity, Claude Code, Cursor, and Windsurf within the same project without losing architectural decisions or project state. The core innovation replaces bloated attention dilution with a curated 4,000-character episodic working memory, dramatically reducing context costs while preserving continuity across model switches.

Empirical testing demonstrates significant value: a 105-session case study on a 50,000-line algorithmic trading system showed zero regressions across 186 unit tests and no context amnesia during model transitions. Production testing revealed 99%+ reduction in reloaded context cost, consolidating over 153,000 tokens of work into under 1,000 tokens (<1% of standard context windows). The architecture features hybrid retrieval combining SQLite FTS5 keyword search with vector embeddings for smart, hot context loading.

The solution offers powerful features including a Save Button for externalizing decisions and constraints, and Smart, Hot Context Loading that maintains context integrity even after session clears. Installation is streamlined through auto-detection of installed environments like Google Antigravity, Claude Code, and Cursor, with explicit setup guides for each platform.