How Clone-Me works

Clone-Me keeps the information and judgment standards that you confirm. While they work, AI agents look up these standards or ask how you would choose. This page explains how memories form and how they are used.

Three promises

How a memory forms

1. 1An agent sends a case
2. 2It waits in the hidden memo box
3. 3The server writes a proposal
4. 4You review it in the inbox
5. 5It becomes a memory

1. Case: When you choose one of several options, correct the AI's suggestion, give a reason, or stress "keep doing it this way", the agent sends that moment as a case. A case holds your exact words and the situation: what the task was, what the AI suggested, and what options there were.
2. Hidden memo box: Cases gather in the server's hidden memo box. Its contents and count do not show on screen. Cases stay for 10 days.
3. Proposal: The server sorts the memo box at most once an hour. It writes a principle only for ways of judging that repeat across conversations or that look important even once, and posts it as a memory proposal. One-time instructions and words that apply only to that task do not become proposals.
4. Review: In the console, go to Inbox → Memory proposals and read each proposal with its supporting cases. Choose one of Accept, Edit, then approve, Only in these cases, Reject or Later.
5. Memory: Only approved proposals become memories. You can edit or withdraw a memory at any time, and each edit is kept as a new version.

Some things become proposals right away. Things you directly asked the agent to remember, information about you (such as your role, work and goals), and things marked as sensitive skip the memo box and go straight to the proposals. You still need to confirm them. Only memories that you write and confirm yourself under Memories → Add directly skip the proposals.

Three groups of memories

Principles and preferences can have conditions (when they apply) and exceptions (when they do not). Agents use such a memory as grounds only when its conditions match.

Dreaming and owner requests

Dreaming is when the server reviews your confirmed memories again, about once a day.

• It does only the very certain work by itself: when a new case supports or contradicts an existing principle, it links the two.
• If anything is even a little unclear, it asks you. That question is an owner request, shown under Inbox → Owner requests.
• An owner request has a question, why it asks, related memories, your words that it is based on, and options. Your answer is applied to your memories as soon as you send it.
• The kinds are support check, new principle, exception, conflict, changed standard, duplicate, outdated information, and scope.

If you do not answer, nothing changes. Nothing is accepted automatically after time passes.

The three tools agents use

Agents use Clone-Me's three tools through MCP, a standard way for AI agents to call tools.

• If ask_me lacks grounds, Clone-Me does not guess. It returns a question for the agent to ask you in the current conversation.
• What you say in the current conversation comes before memories.
• Calls to recall and ask_me show in Call history in the console, with what was asked and what was returned.

Your data

• Memories, proposals, owner requests and call history live only on the Clone-Me server. The website reads them from the server each time it shows a page and does not store them.
• Operators cannot see the content of your memories.
• Under Settings → Data, choose how long to keep raw conversation text and call history. The default is to keep them.
• Cases are sorted with an external AI within the scope you agreed to at sign-up. Cases marked as sensitive are not sent. You can also turn off external AI for each memory.

Glossary

Last updated Oct 28, 2026