Projectmem

What is Projectmem?

Projectmem is a local-first memory and judgment layer for AI coding agents. It captures development history—issues, failed attempts, fixes, decisions, and gotchas—into readable Markdown + JSONL inside your repository, and feeds that context back to your AI agent through the Model Context Protocol (MCP).

How to use Projectmem?

Install with pip install projectmem, then configure your MCP client with command: "pjm-mcp" and args: ["--root", "/absolute/path/to/your/project"]. The server exposes 14 MCP tools that your AI client can use to read and write project memory directly.

Key features of Projectmem

• Persistent project memory across every session
• Pre-commit warnings before repeating a failed approach
• 14 MCP tools for reading and writing memory
• 100% local, no cloud or telemetry
• Works with Claude, Cursor, Antigravity, and Codex
• Reduces context‑rebuilding tokens by up to 50%

Use cases of Projectmem

• Avoid re‑explaining architecture and past attempts to your AI agent
• Get warned at git commit before repeating a known mistake
• Maintain development decisions and gotchas across many sessions
• Reduce token waste from rebuilding context that existed yesterday

FAQ from Projectmem

How is Projectmem different from a simple memory store?

It is not just a memory store—it is a judgment layer that surfaces past failures at the exact moment you would repeat the mistake.

What are the runtime dependencies?

Projectmem requires Python (installed via pip) and a local directory for storage. It is open source (MIT license).

Where does the data live?

All data is stored locally inside your repository as plain‑text Markdown and JSONL files, so you can git diff them.

What transport does Projectmem use?

It uses standard I/O (stdio) via the Model Context Protocol.

Is authentication or cloud needed?

No. Projectmem is 100% local; there is no cloud component or telemetry.