projectmem

Why we included this project

Every coding agent session starts from zero unless something remembers what happened last time. projectmem fills that gap by keeping a plain .projectmem/ directory in your repo where it logs the issues you hit, the fixes that worked, and the decisions you made, then feeds that history back to the agent at the start of each session. The most useful piece is a pre-commit warning that stops the agent before it retries an approach that already failed, which is where most of the wasted tokens go. Everything runs locally with no cloud and no telemetry, and since it ships as a native MCP server plus git hooks, setup is a pip install and one init command. If your team uses Claude Code, Cursor, Codex, or another MCP-capable agent and is tired of re-debugging the same bug every week, this is a low-effort way to give the agent a working memory.