Now in beta
Stop explaining the same things every time. Your AI decides by the criteria you've confirmed, and asks you instead of guessing.
Free during the beta
When a conversation ends, AI forgets what you chose and why. So you keep giving the same explanations and making the same fixes.
TodayEvery new conversation starts with explaining who you are and what you're working on.
With Clone-MeNew conversations pick up your role, goals and current work.
TodayYou've said “lead with the conclusion” a dozen times, and the next conversation forgets it again.
With Clone-MeCriteria you've confirmed are looked up and followed in the next task.
TodayGaps get filled with plausible guesses, and you end up fixing the result.
With Clone-MeWhen there's no basis, it asks you first instead of guessing.
TodaySwitch computers and you teach the same things all over again.
With Clone-MeEvery AI agent you connect uses the same memory, on any computer.
Your AI takes notes as you work, you confirm them, and from then on your AI puts them to use.
Your AI agent notes the moments you choose, correct the AI or give a reason, and the server reads across conversations to write principles for you to review.
Approve, edit or reject proposals in your inbox. Only what you approve becomes memory; anything left alone is never used.
Before it works, your AI agent looks up your criteria and follows them. It shows which memories it used, and asks you when there's no basis.
It refines only what it's sure of and asks whenever anything is unclear.
Anything you haven't confirmed is never used for a judgment, at any step.
Fictional scenarios that show how your clone judges by your criteria. Not real user data.
Tools used: recall
Before writing a report, it looks up the formats you like and dislike.
You: Write this week's status report.
Checked with your clone: recall2 foundLead with the conclusion
Long-winded prose
AI agent
Conclusion first, and the body trimmed to three paragraphs.
Tools used: ask_me
Before starting a large change, your AI agent gets its plan checked first.
You: Rewrite the payment module from scratch.
Checked with your clone: ask_meNeeds changesReview a step-by-step plan before big implementations
Show the scope, order and done criteria first.
AI agent
Here's the plan first. Step 1 order model, step 2 payment provider, step 3 webhooks…
From the tools that connect your AI agent to the console where you review and manage memories. Features still in progress are marked.
Connect an AI agent such as Claude Code with a single connection token. The agent uses three MCP tools to save memories, look them up and ask for judgments.
rememberSave memoriesrecallLook up memoriesask_meAsk for a judgmentSee the proposals the server raised from the cases your AI agent left, and approve, edit or reject them in one place. Nothing becomes memory unless you choose it.
Browse confirmed memories by type, and edit, archive or withdraw them at any time.
Create a separate connection for each computer and AI agent, and issue, reissue or revoke connection tokens.
Periodically reviews stored information, decision cases and principles to keep your memory current. It does not train the AI model itself.
It remembers only what you confirm, shows its evidence, and asks when it doesn't know. A judgment is never permission to act.
Neither facts you stated nor preferences the AI inferred are used for judgments until you confirm them. No answer is not approval.
Every judgment comes with the memories and versions it used, and explains why it holds off when the evidence is weak. Linking a judgment back to the cases and conditions a principle came from is planned.
When nothing it knows would change the judgment, it asks you instead of guessing.
One account, one memory space. Your gaps are never filled with someone else's similar cases.
Likes and dislikes are kept with the conditions and exceptions for when they apply. A single correction never spreads to every situation.
Dreaming applies changes automatically only when evidence under the same conditions adds to a criterion you confirmed. New interpretations need your confirmation. Repeats don't add evidence. Anything even slightly unclear comes to you as an owner request.
Your clone's answer is advice for your AI agent. It does not approve real-world actions such as sending an email or making a payment.
Editing, withdrawing and archiving memories is always up to you. Data export and account deletion are being prepared so you can do them yourself too.
Nothing is used before you confirm it, and you can edit or disconnect at any time.
One account, one memory space. Your gaps are never filled with someone else's cases.
Proposals you haven't approved are never used for lookup, judgment or reinforcement.
The admin screens have no way to open memory content. Issues are traced by request ID.
Edit or withdraw memories and disconnect agents yourself. Data export and account deletion are in progress.
Plans and prices will be published after the beta, and we'll let you know before anything becomes paid.
Freeduring the beta
The plan for invited users during the beta. When the beta ends we'll publish paid plans and let you know before anything changes.
Up to 10 connections, up to 5 active tokens per connection
Have another question? Send it through the contact form.
Sign-up is invite-only for now. Leave your email and we'll send an invite as soon as a spot opens.