Start with your next session.

No. Mnemolith runs entirely on your machine: the engine only accepts connections from your own computer, and your memories live in ~/.mnemolith. The only things it downloads are, once, the embedding model it uses to search memories by meaning (about 90 MB) and, if you don’t have Java, a private copy of Java. Your AI tool still talks to its own provider the way it always does; Mnemolith doesn’t add anything to that.

mnemolith setup connects Claude Code, Cursor, and Codex automatically, for all your projects. Any other tool that supports MCP works too: run mnemolith connect --print for the settings to paste into it.

Most people should use the one-line installer at the top of this page. It installs everything it needs (uv, and through it Python), installs Mnemolith, and runs setup.

If you already manage Python tools yourself, use what you know: uv tool install mnemolith, pipx install mnemolith, or pip install mnemolith (Python 3.10 or newer), then run mnemolith setup. They all install the same version, and you don’t need Java for any of them.

No. The engine runs on Java 11 or newer. If you already have it, Mnemolith uses yours. If you don’t, Mnemolith downloads a private copy (about 50 MB) into ~/.mnemolith/java the first time it needs one, checks it against its published checksum, and uses it only for itself. Nothing is installed system-wide, and uninstalling removes it.

When your AI saves a memory, it can link it to the files the fact depends on, like docker-compose.yml. If one of those files changes, the memory is marked stale: your AI stops seeing it, and the dashboard (mnemolith ui) asks you to decide. Click Still true to keep it, or Retire if it’s wrong now.

Notes files and built-in memories never know when they’ve gone out of date, and each tool keeps its own. Mnemolith is one memory shared by all your AI tools, and it checks memories against your files so outdated ones stop being used.

Already have notes? Run mnemolith import, or click Import notes in the dashboard, to turn your CLAUDE.md, AGENTS.md, and Cursor rules into memories. You review them before anything is saved.

Yes. Memories are kept per project, named after the project’s folder, and change detection uses file fingerprints, so git isn’t required.

It’s built for macOS, Windows, and Linux. So far it has been tested most on macOS, so if something goes wrong on Windows or Linux, please tell me using the form below.

Not yet: each computer keeps its own memory. To share today, run mnemolith sync, which writes the current memories into your repo’s AGENTS.md and CLAUDE.md so they’re committed with the code. Shared team memory is planned.

Run mnemolith export, or click Export in the dashboard. You get a JSON file with every memory, the files it’s linked to, and its full history.

Mnemolith used to be called Aether Cache. Run the install command above: it installs Mnemolith, removes aether-cache, moves your memories over, and reconnects your AI tools. Nothing is lost.

Run mnemolith uninstall. It disconnects your AI tools, stops the engine, removes the private Java copy if there is one, and asks before deleting your memories. Then remove the program with uv tool uninstall mnemolith.

Yes. There’s no account and no paid plan, and the code is on GitHub.

Run mnemolith doctor: it checks everything and says how to fix what it finds. If the mnemolith command isn’t found after installing, run uv tool update-shell and open a new terminal. Still stuck? Send the output of mnemolith doctor with the form below.