Memory for Claude Code
Claude forgets.
Cogi doesn't.
Cogi the squirrel keeps every message you and Claude say, like nuts in its hollow. When something drops out of Claude's memory, Cogi brings it back.
Free · runs on your computer · nothing leaves it
How it works
Collect
After every answer, Cogi picks up the new messages. You do nothing.
Keep
Each message is stored word for word, with its date, in your hollow: a folder on your computer.
Find
When Claude needs something from earlier, Cogi hands back the few messages that matter.
Tested, not promised
As good as rereading the whole conversation
On the LoCoMo benchmark, answers from the 1,000 tokens Cogi hands back score 74.7%. Handing over the whole conversation, twenty times more text, scores 76.5%.
answers right, scored the way Mem0 and Zep were
to search 10 million tokens, on a laptop CPU
per message: no account, no API key, no AI calls to save
Why Cogi
Private
Your hollow is a folder on your computer. No cloud, no account.
Exact
You get the original words and when they were said. Cogi never makes things up.
Automatic
It saves after each answer and hands Claude what's relevant with each prompt.
Get started
One command in a terminal, with uv installed:
claude plugin install cogito-mem --marketplace meridian21lab/cogito-memThe full guide, with what you should see at each step →
Where Cogi works
| App | Works? | What you get |
|---|---|---|
| Claude Code, in a terminal or an IDE | Yes | Everything: saves and brings things back on its own |
| Claude Code in the desktop app | Yes | The same plugin (we are still testing it there) |
| Codex | Yes | Saves every turn; searches when asked (one-line installer) |
| Claude Desktop chat | Partly | Save and search tools, but no automatic saving |
| Cloud sessions, claude.ai | Not yet | On the roadmap |