A new open-source project called Engrim has launched on GitHub, offering what its creators describe as a universal, cross-model episodic memory standard for AI coding agents. The tool provides persistent memory via a local-first SQLite engine, meaning all data stays on the developer machine with no cloud dependency whatsoever. This design choice makes it particularly appealing for teams working with sensitive or proprietary codebases.
Engrim is designed to work across multiple AI coding assistants, including Google Antigravity, Anthropic Claude Code, Cursor, and Windsurf, giving each agent the ability to recall context from previous sessions, projects, and conversations. This addresses one of the key limitations of current AI coding tools: their inability to remember what happened in prior interactions. Developers frequently find themselves re-explaining the same architectural decisions and coding patterns every time they start a new session.
The project is project-scoped, meaning memory is organized around specific codebases rather than a global knowledge store. This approach mirrors how developers actually work, thinking in terms of projects, repositories, and codebases, rather than forcing a one-size-fits-all memory model. Each project gets its own isolated memory space, preventing cross-contamination between unrelated codebases.
While still early with just 14 GitHub stars at launch, Engrim tackles a real pain point that has become increasingly frustrating as AI coding agents grow more capable. The lack of persistent memory across sessions means valuable context is lost every time a developer closes their editor. A local-first approach also appeals to developers concerned about sending proprietary code context to third-party servers, a growing concern in enterprise environments where data sovereignty matters.