Engrim：本地优先的SQLite内存引擎助力AI CLI跨模型协作

开发者Tim Gordon推出了Engrim，一款开源的本地优先SQLite内存引擎，旨在解决开发者在使用不同AI模型和开发环境时的上下文管理问题。1该工具支持Google Antigravity、Claude Code、Cursor MCP、Windsurf和Codex CLI等多个AI环境，允许开发者在项目中自由切换模型而不丢失架构决策和项目状态。1

Engrim的核心优势在于其高效的上下文压缩能力。1该工具成功将153000个token的工作压缩至不足1000个token的活跃记忆包中，实现99%以上的上下文重载成本削减。1在实际测试中，Engrim在包含50000行代码的算法交易系统上完成了105个连续会话，零回归零漂移，并通过了186个单元测试。1内存数据存储在本地的~/.engrim/memory.db文件中，采用SQLite FTS5和model2vec混合检索机制。1

该项目采用MIT许可证发布，开发者可通过pip install engrim命令安装，安装过程中工具会自动检测已安装的AI环境并配置所有模块。1

Tim Gordon has released Engrim, an open-source SQLite-based memory engine designed to enable developers to switch between different AI models and environments without losing project context or architectural decisions 1. The tool supports multiple AI platforms including Google Antigravity, Claude Code, Cursor MCP, Windsurf, and Codex CLI, allowing teams to work across environments while maintaining continuity 1.

The engine achieves significant efficiency gains by compressing 153,000 tokens of project work into fewer than 1,000 tokens of active memory, reducing context reload costs by over 99 percent 1. Testing on a 50,000-line algorithmic trading system demonstrated the tool's reliability through 105 consecutive sessions with zero regressions and 186 unit tests with zero failures, with no performance drift when switching between models 1. Engrim stores memory locally at ~/.engrim/memory.db using a hybrid retrieval approach combining SQLite's full-text search capabilities with model2vec indexing 1.

Installation is straightforward through pip install engrim, after which the tool automatically detects installed environments and configures all necessary modules 1. The project is released under the MIT license 1.