Why does my AI forget everything I told it?

Your AI forgets because each session ends and the next one starts empty, and Magnemo gives that next session a record you approved to start from.

What is actually happening

An AI works inside a window of text, and that window is all it has. When a long session fills the window, the AI compacts: it boils the work down to a summary and guesses at what mattered. When the session ends, even the summary goes. The next session opens with nothing, so you explain the project again.

That is not your fault, and no prompt fixes it. The model cannot hold your project between sessions.

What a record you approved changes

Magnemo is an open-source memory that runs on your own computer. Your AI can propose a note, but nothing is kept until you say yes. Each kept note carries who wrote it, who approved it, and when. A newer note replaces an older one, and the older one is archived with its words intact.

At the start of the next session, the AI reads that approved record first. It begins from what you confirmed, not from a guess.

The memory is a folder of plain markdown files. You can open it in any editor.

The paste and the config

Install it, then make a vault, which is the folder that holds the memory:

pip install magnemo
magnemo init ./vault

Add this to the config for Claude Desktop or Cursor, with the path to your own vault:

{"mcpServers":{"magnemo":{"command":"magnemo-mcp","env":{"MAGNEMO_VAULT":"/absolute/path/to/vault"}}}}

It also works with Claude Code and any other MCP client. When your AI proposes a note, you approve it with magnemo yes. Your AI cannot approve its own notes.