projectmem

Overview of projectmem

What it is

projectmem is an open-source agent memory layer for AI coding agents. It records issues, attempts, fixes, and decisions made during a project’s development and stores them locally in a .projectmem/ directory inside the repository. A native MCP server provides 17 tools to clients such as Claude Code, Claude Desktop, Cursor, Antigravity, and Codex, allowing the agent to read and write memory autonomously. The system is local-first with no cloud, account, or telemetry. It is released under the MIT license.

Key features

• Stores typed events (issues, attempts, fixes, decisions, notes) in a local, event-sourced log
• Provides a pre-commit warning that fires before an approach that already failed in the project is repeated
• Offers a native MCP server that integrates with Claude Code, Cursor, Antigravity, and Codex
• Includes a pjm doctor command that finds unregistered projects and checks configuration
• Supports cross-project memory inheritance, git hooks for pre-commit, post-commit classification, and post-merge tracking
• Generates a local visualization (pjm visualize) from .projectmem/events.jsonl showing solved cases and failed approaches

Use cases

Coding teams or individual developers using AI coding agents who want the agent to remember past debugging attempts, decisions, and gotchas across sessions. Useful for projects where agents burn tokens re-reading files each session and repeat failed approaches. Also applicable in any environment where a stateless AI model needs a persistent memory layer.

Getting started

bashpip install -U projectmemFind projects:bashpjm doctorRegister them:bashpjm doctor --fixConfigure the MCP server in your client:json”mcpServers”: { “projectmem”: { “command”: “/absolute/path/to/python”, “args”: [“-m”, “projectmem.mcp_server”] }}Initialize in a project:bashpjm init

License and pricing

Licensed under MIT. Open source, free to use. Pricing is not stated as it is open source.

這是什麼

projectmem 是一個開放原始碼的 AI 編碼代理記憶層。它記錄專案開發過程中遇到的問題、嘗試的方法、修復的內容與做出的決策，並將這些資料存放在儲存庫內的本機 .projectmem/ 目錄中。一個原生的 MCP 伺服器提供 17 個工具給 Claude Code、Claude Desktop、Cursor、Antigravity 與 Codex 等客戶端，讓代理能自主讀取與寫入記憶。系統採本地優先設計，無雲端、無帳號、無遙測。採用 MIT 授權。

主要功能

• 儲存類別化事件（問題、嘗試、修復、決策、筆記）於本機事件溯源日誌中
• 在重複已失敗方法之前觸發提交前警告
• 內建 MCP 伺服器，可與 Claude Code、Cursor、Antigravity 與 Codex 整合
• 提供 pjm doctor 指令，搜尋未註冊的專案並檢查設定
• 支援跨專案記憶繼承、提交前鉤子、提交後分類與合併後追蹤的 git 鉤子
• 從 .projectmem/events.jsonl 產生本機視覺化圖表（pjm visualize），顯示已解決案例與失敗方法

使用情境

使用 AI 編碼代理的團隊或個別開發者，希望代理能記住過去的除錯嘗試、決策與注意事項，避免每次會話都重新讀取檔案並重複失敗的方法。適用於當無狀態 AI 模型需要一個持續的記憶層時。

快速開始

bashpip install -U projectmem尋找專案：bashpjm doctor註冊專案：bashpjm doctor --fix在客戶端中設定 MCP 伺服器：json”mcpServers”: { “projectmem”: { “command”: “/absolute/path/to/python”, “args”: [“-m”, “projectmem.mcp_server”] }}在專案中初始化：bashpjm init

授權與價格

採用 MIT 授權。開放原始碼，免費使用。價格未說明，因其為開放原始碼。

Based on the project's public README and repository metadata.