How Clone works.
AI agents do real work on their own, then keep stopping to ask you. Your time becomes the limit. Clone learns how you work and answers for you, so only the questions that really need you reach you.
Layer 1 · Recording
We learn how you actually work.
Clone records what you do, how you talk to your AI tools, and the apps you use. All in real time.
What you do on your computer
Screen, typing, clicks, and app switches as they happen. Not pieced together later from logs.
How you work with AI
Every prompt and reply across Claude Code, Codex, and the rest, saved next to the work itself.
The tools you already use
GitHub, Calendar, Slack, Notion, Gmail, Drive, voice. Recorded on your computer first. You choose what uploads.
Layer 2 · Memory
A memory that grows with you.
Two memories. One for what happened, one for what you tend to do. Both keep improving without losing the older you.
What happened, and when
Every moment timestamped. Replay a session, jump to the exact step, or pull context from days ago.
What you tend to do
Patterns across sessions. How you name things, the order you do them, the shortcuts you trust.
Learns more without overwriting
New memory lands without erasing old. Turn off any source you don't want remembered.
Layer 3 · Prediction
Skip the prompt. Clone's got it.
Clone guesses what you'd say to your agent. If confident, it says it. If not, it asks you first.
Guesses what you'd say next
Clone has a guess and a sense of how sure. Confident ones go to the agent, the rest come to you.
Careful before fast
Better to send fewer answers and get them right. A wrong reply costs more time than a missed one.
Two places to use it
Clone CLI in your terminal. Clone in every other app.