Digital twin
Your digital twin
The twin learns, from your conversations, how you decide, how you phrase things and where you lean. The better it knows you, the fewer confirmations you need before it stays on your intent.
Where to find it
Open More → Agents in the sidebar and your twin is sitting at the top of the list, under You. It opens in the same screen as your role agents, so you look inside it and grow it exactly the same way.
Learning can also be switched on or off, and its state checked, from Settings → Self-agent.
Turning it on
Open ⚙ Settings in the twin's screen.
Switch on Learn automatically from conversations.
While it's on, personality — judgement, communication, strategy — is learned from conversation and accumulated into a profile. Switching it off stops learning, and the profile built so far stays usable.
What it learns
Learning is organised along six axes.
| Axis | What it watches |
|---|---|
| Judgement | What you prioritise and what you're willing to drop |
| Communication | How you phrase things, your tone and distance |
| Strategic stance | Whether you push or protect, and where you place your bets |
| Turns of phrase | The wording you reach for |
| Interests and domain | Where your time actually goes |
| Track record | Decisions you've made and how they turned out |
Each axis carries a confidence level — high, medium, low or none — displayed on the Personality map.
Reading the maturity level
Maturity reflects how much has been learned from conversation. It moves through these stages:
| Stage | Meaning |
|---|---|
| Not yet observed | Nothing learned so far |
| Emerging | Fragments are starting to show |
| Taking shape | Patterns are becoming visible |
| Well understood (still updating) | Captured from several angles. This is the top of the scale |
As the app puts it, the twin is never “finished”. People keep changing, so what's shown is not a completion percentage but how richly and how recently you've been captured — thickness, breadth and freshness — along with the trajectory over time. Even at its best, it reads “well understood (still updating)”.
Growth rings
The rings record when each facet developed. Once enough snapshots exist you can overlay “before versus now”, which makes it clear what has shifted recently and what has settled.
The observation log shows what triggered each piece of learning. Early on it simply reads Nothing yet. These build up as you talk.
Where it takes effect
Once grown, the twin carries your judgement and your way of putting things into the AI in three places.
| Situation | Whether it applies |
|---|---|
| Ordinary chat (no role) | When Apply your twin in ordinary chats too is on in ⚙ settings (on by default) |
| While a role agent is in use | Whatever that role's ⚙ settings say under Overlay your persona |
| Auto mode | Always — the twin is the one replying |
Observations picked up from conversation don't reach the AI as they are. They start shaping replies only once you adopt them from the observation log into the profile.
The observation log, and adopting from it
Your decisions and turns of phrase are picked up automatically from everyday conversation and collect in the observation log. Per axis, the twin's screen shows N observed alongside N new for the ones you haven't adopted yet.
Press Summarise N observations.
It reads what has collected and suggests what is worth adding to your profile.
Only what you Adopt enters the profile, and it applies from the next conversation on.
That means what you typed, sends you queued, instructions from the phone remote, and your answers to confirmation cards. An autonomous run's automatic continuations, the stages it executes and the text that kicks off a loop are assembled by the machine, so nothing is learned from them.
Talk to your twin
The right-hand side of the twin's screen is a chat. There are three ways to use it.
| Use | Example | What happens |
|---|---|---|
| Teach it | “I put continuity ahead of short-term revenue” | A proposal appears for the matching axis. Whether it's adopted is up to you |
| Test it | “How would you call this situation?” | The twin answers from the current profile, naming in the reply the judgement axes it used, so you can see how far it has come |
| Hand it material | Attach a CV or a written summary of your thinking with 📎 | It reads it, then proposes what's worth adding to the profile |
Materials
Drop documents about yourself — a CV, your track record, your thinking written down — onto the Materials panel and they're stored in a knowledge base called Your materials, converted to OKF so they can be searched and referenced. The twin draws on them as grounds for its decisions. Role agents' materials work the same way (Growing your agents).
Auto mode
Choosing Start auto mode in the composer lets the twin take over the decisions and keep going. Stop auto mode ends it.
Who replies, in the 🦝 button's ⇧ menu, changes who handles the automatic replies. The default is your twin, but appoint the reviewer — or an agent you built yourself — and it answers on your behalf with that role's judgement and voice. The reply bubble is labelled 🎭 Auto-reply by …, so who answered is always visible.
Stop auto mode halts the twin's automatic replies, but a response already being generated still finishes on screen. To cut off the response itself, press Esc.
This is not the same as Run to completion (work mode), which resumes a task that ended midway. The difference is explained in Let it run on its own.
Storage and privacy
The learned profile is stored as a plain Markdown file at identity/self-profile.md inside your save folder. Anything you add or remove in the twin's screen writes to that file, and you can edit it directly in any text editor — or in Obsidian, if you have opened the save folder as a vault. You can find or change the save folder under Settings → Storage, export and deletion.
Your twin is you, so it is left out of handing a role over. It cannot be exported.
Where to go next
- Let it run on its own — work mode versus loops
- AI team — splitting work across roles
- Growing your agents — growing each role's own contents (a separate layer from the twin)
- Settings and where data lives — where things are stored