Patdolitse/piia-engram

View on GitHub ↗

Local-first AI memory you can see, edit, and override — portable across Claude Code, Codex, Cursor, Windsurf, and other MCP coding tools.

157 ★11 forksPythonUpdated 18d ago

What you need to know

Local-first AI work identity layer - tell AI once who you are and how you work, portable across MCP coding tools like Claude Code, Codex and Cursor

Install

pip install piia-engram && engram setup

Usage

  • Run engram setup to detect AI tools and write the MCP connection
  • Start sessions with get_user_context, store lessons and decisions, and search knowledge across tools

Key features

  • 17 core MCP tools plus 40 advanced opt-in tools
  • Local JSON storage under ~/.engram that you own and can edit
  • User approval for high-risk writes with a strict-mode option
  • Cross-tool continuity across Claude Code, Codex, Cursor and more
  • Knowledge inheritance across projects
  • Passive knowledge capture and automatic playbook extraction
  • Optional hybrid search and remote deployment over SSE

Best for

Developers using multiple AI coding tools who don't want to re-explain themselves every session

Caveats

  • AGPL-3.0 license
  • Local files are plaintext JSON by default - optional field-level encryption
  • Stores identity and lessons, not agent task-memory or session history
Platforms: macOS · Linux · WindowsClients: Claude Code · Codex · Cursor · Claude Desktop · Hermes · Windsurf · Any MCP client

Reviewed 2026-08-11

Topics

agent-memoryai-agentai-identityai-memoryanthropicclaude-codeclaude-desktopcodexcontext-managementcursorllm-toolslocal-firstlong-term-memorymcpmcp-servermemorymodel-context-protocolpythonvibe-codingwindsurf
Stars
157★
Forks
11
Language
Python
License
AGPL-3.0
Created
2026-05-18
Last push
2026-08-17