Memory for AI agents
Your AI agents stop forgetting.
Every new session starts from zero, so you re-explain your stack, your preferences and the bug you already fixed. gitmind gives Claude, ChatGPT, Claude Code and Codex one shared memory: agents save what they learn as they work and recall it next time, in whichever app you use. It's open source, and your memory is Markdown notes in a git repository you own.
# claude.ai, ChatGPT: add a connector with this URL
https://gitmind.app/mcp
# Claude Code
claude mcp add --transport http gitmind https://gitmind.app/mcp
# or keep a synced clone on your laptop
curl -fsSL https://gitmind.app/install.sh | sh
gitmind setup
Features
Works in every AI app
One MCP endpoint for claude.ai on web, desktop and mobile, ChatGPT, Claude Code and Codex. What one app learns, the others know.
Automatic
Agents save decisions, preferences, gotchas and setup knowledge as short facts while they work, and load them when the next session starts. Nothing to approve.
Facts stay current
When something changes, the old fact moves to a dated History instead of contradicting the new one, and agents are warned when they save something that conflicts.
Your memory is a git repo
Every note is Markdown, every change is a commit. Clone it, read it in Obsidian, see its history, take it with you.
Shared with your team
Team workspaces share a memory between people and their agents, with invite links and roles.
Open source
AGPL-3.0. Use gitmind.app, or self-host it on your own Cloudflare account in a few minutes.
How it works
- Connect. Sign in with GitHub and add gitmind to your AI apps. Each one signs in to your memory once.
- Work as usual. At the start of a session the agent loads what's relevant: pinned notes and an index of the project's notes. It searches when it needs more.
- Keep what matters. When something is worth remembering, the agent writes it down as a short fact. Next time, in any app or on any machine, it already knows.
Questions
What is MCP memory?
MCP (Model Context Protocol) is the open standard AI apps use to call outside tools. An MCP memory server gives an agent tools to search, read and save notes, so it remembers things between conversations. gitmind is one: the agent decides what to save, and gitmind stores, indexes and syncs it. See the tools it provides.
How do I give Claude Code persistent memory?
Run claude mcp add --transport http --scope user gitmind https://gitmind.app/mcp, or install the gitmind CLI and run gitmind setup. The CLI keeps a synced local clone of your notes and adds hooks that load a project's notes at the start of every session and remind the agent to save what it learned. See CLI and local sync.
Can Claude and ChatGPT share the same memory?
Yes. Add https://gitmind.app/mcp as a custom connector in claude.ai and in ChatGPT, and sign in to the same account in both. They read and write the same notes, as do Claude Code and Codex. See Connect your AI apps.
Is my memory private?
Your personal workspace is readable only by you; team workspaces only by their members. Every app signs in with OAuth after a consent page, gitmind runs no LLM and never sees your conversations, only what an agent chooses to save or search, and it refuses to store notes that look like secrets. See Security and privacy.
Can I self-host gitmind?
Yes. gitmind is open source under AGPL-3.0 and deploys to your own Cloudflare account (Workers Paid plan) with a few commands. Or skip the server entirely: the CLI can sync notes through a private GitHub repository. See Self-hosting.
How is gitmind different from ChatGPT or Claude's built-in memory?
Built-in memory belongs to one app and is managed by its provider. gitmind is one memory shared by every AI app and machine you use, stored as Markdown files in git that you can read, edit, version, export and share with a team. See how gitmind compares.
Is gitmind free?
Yes. gitmind.app is free to use with a GitHub sign-in, and the code is open source.