Projectmem

projectmem is an open source memory layer for AI coding agents. It records the issues, attempted approaches, fixes and decisions that arise while a project is built, stores them in the repository, and feeds that history back to the agent at the start of each session. The aim is to stop the agent re reading the codebase from scratch and to warn it before it repeats an approach that has already failed. It works with Claude Code, Cursor, Codex, Antigravity, Grok and any MCP client, and with local coding models such as DeepSeek, GLM and Qwen through a universal mode. A single MCP server serves every registered project. Setup is a pip install followed by a command line tool, with commands such as pjm doctor to find existing projects and check client configuration, and pjm init to enable memory in a repository. The vendor lists 25 CLI commands and 17 MCP tools, and links a research paper on arXiv. The tool is aimed at developers who use several coding agents and lose context between sessions. Everything runs locally. The vendor states that it is MIT licensed, collects no telemetry, and has no cloud service or subscription, so there is no cost to use it. The site claims savings of up to 50 percent or more of AI tokens, a figure given by the vendor rather than independently measured.