Repolode
Submit
timgordontg

timgordontg/engrim

A local-first, project-scoped SQLite memory engine that shares coding decisions and project state across AI coding agents like Claude Code, Cursor, and Codex CLI.

View on GitHubWebsite
Stars
304
Forks
16
Open issues
4
Last commit
last month

What is it?

engrim is a Python tool that gives AI coding agents a shared, persistent memory for a codebase. Instead of replaying long chat transcripts, it stores curated records such as architectural decisions, facts, constraints and the current task in a local SQLite file at ~/.engrim/memory.db, then injects a size-capped memory pack when a new session starts.

It integrates with Google Antigravity, Claude Code, Cursor, Windsurf, Codex CLI, GitHub Copilot CLI and OpenCode, using lifecycle hooks where the agent supports them and an MCP server (engrim serve --mcp) otherwise. Records are tagged with the agent that wrote them, and recall combines FTS5 full-text search with static vector embeddings from model2vec, which can be disabled for pure lexical search. The engine also provides a health check (engrim doctor) that verifies database integrity and hook configuration, plus CLI commands for backup, merge and pruning.

All data stays on the local machine with no telemetry or cloud sync, and the database is created with owner-only permissions.

Key features

  • 01Shares one local SQLite memory store across several AI coding agents on the same project.
  • 02Injects a size-capped memory pack at session start, with a resume pointer for unfinished work.
  • 03Hybrid recall combining FTS5 full-text search and model2vec vector embeddings, with pure-lexical mode available.
  • 04Records the originating agent for each memory entry.
  • 05Provides engrim doctor to check database integrity and hook configuration, with an auto-fix option.
  • 06Exposes engrim_* tools over a stdio MCP server, and includes backup, merge and prune commands.

Use cases

  • Continuing a codebase in Claude Code or Cursor after clearing the chat, without re-explaining architecture decisions.
  • Switching between several AI coding agents on the same repository while keeping one shared decision log.
  • Keeping project memory local for codebases where sending context to third-party memory services is not acceptable.
  • Carrying memory across CI runs by merging exported SQLite stores.

Quick start

  1. 1

    Install engrim from PyPI

    pip install engrim
  2. 2

    Auto-detect and configure installed agent environments

    engrim setup
  3. 3

    Check the installation health

    engrim doctor

Things to know

  • Requires Python 3.10 or newer.
  • Stores memory in a local SQLite file at ~/.engrim/memory.db, with no cloud sync between machines.
  • Codex CLI hooks must be reviewed and trusted manually before they run.

In collections

Long-term memory for AI agentsMCP servers for Claude CodeClaude Code tools, skills & pluginsOpen-source AI agent frameworks & appsAI coding assistants & agentsMCP serversPrivacy & local-first toolsDatabases & SQL tools

Topics

#ai-agents#local-first#sqlite#memory#mcp#developer-tools#claude-code#cursor#antigravity#codex

FAQ

What is engrim?
A local-first, project-scoped SQLite memory engine that shares coding decisions and project state across AI coding agents like Claude Code, Cursor, and Codex CLI. engrim is a Python tool that gives AI coding agents a shared, persistent memory for a codebase. Instead of replaying long chat transcripts, it stores curated records such as architectural decisions, facts, constraints and the current task in a local SQLite file at ~/.engrim/memory.db, then injects a size-capped memory pack when a new session starts.
Is engrim free to use? What license does it have?
engrim is open source under the MIT license. Check the license terms before using it in commercial products.
Is engrim still maintained?
Updated within 6 months. The last commit to engrim was on Sep 26, 2026.
What can I use engrim for?
Continuing a codebase in Claude Code or Cursor after clearing the chat, without re-explaining architecture decisions. Switching between several AI coding agents on the same repository while keeping one shared decision log. Keeping project memory local for codebases where sending context to third-party memory services is not acceptable. Carrying memory across CI runs by merging exported SQLite stores.
How do I install engrim?
Install engrim from PyPI: pip install engrim Auto-detect and configure installed agent environments: engrim setup Check the installation health: engrim doctor
Where can I get engrim?
The source code and installation instructions are on GitHub: https://github.com/timgordontg/engrim

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects