MAGNEMO is Persistent Memory for AI Agents.
Kept on your computer, in plain files.
Your AI writes down what it learns.
You approve what matters.
Give your agent a notebook. Install in one paste.
Paste this to your agent. Claude Code, Cursor, Claude Desktop.
Get the notes when this changes: each release, and each new app MAGNEMO connects to.
Your Memory lives on your computer, in plain files you can open. No network calls of its own. Add a backup remote and it pushes there, nowhere else.
Works with any AI that speaks MCP. One paste for Claude Code; a config block for the rest. Claude Code, Cursor, Claude Desktop, and the rest.
A real terminal, real machine time: the install, the vault, the mount — then the session is killed, and a brand-new one resumes from Memory, quoting the receipt it resumed from.
Every Agent needs a Memory; the only question is whether it’s one you can trust.
Every memory tool remembers on day one. Every one of them. Every demo looks the same.
One wrong thing gets in: a bad instruction, a guess, a note nobody checked. It’s remembered as fact. Now the memory is lying to you, and you can’t tell which part.
Nobody trusts it, so everyone explains everything again. The memory has turned into dead weight, a hoard nobody trusts.
That’s Memory rot. It happens to every memory that keeps everything. MAGNEMO keeps only what you approve, with a receipt on every entry. On day 500, you still believe it.
Your AI plugs into MCP like USB-C plugs into your device. The one plug standard. MAGNEMO: The one plug Memory.
pip install magnemo · magnemo init · magnemo mount writes .mcp.json, the file Claude Code reads, and the first line in CLAUDE.md.
Four verbs for memory, three for your yes, two for a save point. Plain files in a folder you can open.
For builders: Through the MCP server there is no write path except a draft. Every entry carries who wrote it, when, and who approved it. Trust is earned on the record and lost on a failure, never flipped by a switch.
Your Agent wakes from its own record. Every decision still decided. Nothing forgotten.
Your AI can build in an afternoon what used to take a month. It can also get things wrong just as fast, and build on top of them. Speed is no longer the hard part. Knowing what’s true is. The faster you build, the more expensive one wrong memory gets. MAGNEMO is the Memory your AI keeps: only what you approve, and each one shows who approved it.
Each release, and each new app MAGNEMO connects to.