projectmem

Local-first memory and judgment layer for AI coding agents.

Every new AI session starts from zero — re-explaining the architecture, re-trying fixes that already failed, burning thousands of tokens rebuilding context that existed yesterday. projectmem fixes that. It captures your development history — issues, failed attempts, fixes, decisions, and gotchas — into readable Markdown + JSONL inside your repo, and feeds it back to your AI agent through the Model Context Protocol.

Key features

• Persistent project memory — captures issues, attempts, fixes, decisions, and gotchas across every session
• Pre-commit warnings — warns you at git commit before you repeat a failed approach
• 14 MCP tools — your AI client reads and writes memory directly
• Local-first — 100% on your machine, no cloud, no telemetry; plain-text storage you can git diff
• Cross-client — verified with Claude, Cursor, Antigravity, and Codex
• Token-efficient — cuts context-rebuilding tokens by up to 50%

Installation

pip install projectmem

MCP configuration

{
  "mcpServers": {
    "projectmem": {
      "command": "pjm-mcp",
      "args": ["--root", "/absolute/path/to/your/project"]
    }
  }
}

Why projectmem

It's not just a memory store — it's a judgment layer. projectmem remembers what went wrong and surfaces it at the exact moment you'd repeat the mistake. Open source, MIT-licensed.

• GitHub: https://github.com/riponcm/projectmem
• PyPI: https://pypi.org/project/projectmem/