The three kinds of twin

One engine, three audiences. They share the same grounding, the same citations and the same honest "I don't know" — what differs is who they answer and who is allowed to reach them.

Your public twin#

You, answering your audience, at hiy.ai/yourname. Built from what you've actually made — writing, talks, notes, a CV. It's the one you share, embed, and put on a business card.

Before it goes live it passes an identity check, because a public twin claims to be a specific person. A cleared twin carries a Verified badge beside its AI label.

The badge is tied to the name that was checked. Rename your twin away from that name and the badge comes off — a check made for one name isn't carried over to another. Your twin stays live and keeps answering throughout; renaming back restores the badge on its own, with no second review.

Support Agents#

An agent for a product you own. It answers your users from your product's material, in a neutral voice — it represents the product, not a person.

It's reachable only through a token your own site mints. Anything it can't answer goes to your inbox with the visitor's email attached, so a stuck user never dead-ends.

Team Agents#

You, answering colleagues from internal material. Invites are just email addresses, and access begins at first sign-in. Everyone else — search engines included — gets a 404 rather than a login wall, so the agent's existence isn't advertised.

A Team Agent reports what you wrote. It never decides on anyone's behalf, and there's deliberately no per-person trail.

Switching between them#

Twins live behind the twin switcher in your dashboard. Each keeps its own knowledge, its own audience and its own rules — adding material to one never affects another.

You may only twin yourself, or something you own. What enforces that isn't your address being unique — it's the identity check every public twin passes before going live, a person reviewing anything the automated check can't clear, and a report link on every public page that no plan can remove.

There's a ceiling on the other two kinds: three Support and Team Agents per organisation, counted together, alongside your own public twin. Ask for a fourth and you're told so before anything is written — it's a limit we hold for now rather than one an upgrade lifts.

An unfinished agent still holds a slot#

A slot is spent the moment an agent exists, not when it goes live. An agent you started and left — still a draft, with no sources yet — counts against the three exactly as a live one does. So a couple of abandoned tabs can be the whole reason a fourth one won't be made.

That count reads one thing: whether the agent has any sources on it. It doesn't look at your Lists — so an agent you've only given rows to still reads as unfinished, and those rows are all still there when you open it.

Nothing is swept up for you, and that's deliberate: a draft is a real, trainable agent, and one you unpublished on purpose looks identical to one you walked away from. Guessing between them would delete somebody's work. So the refusal says the true thing instead — an unfinished agent still holds a slot — and leaves the choice with you: open that agent from your agents list, delete it, and the slot is free again. Deleting an agent covers what goes with it and what doesn't.

Only the account that created an agent can delete it, and the refusal now says so. Slots are counted across the whole account, so if you were invited into somebody else's account you can hit the cap on a draft one of your colleagues abandoned — and there is nothing you can do about that one. When that happens the refusal names the agent and says the account that created it has to free the slot, instead of pointing you at a Delete button you do not have. If the abandoned draft is your own, you get the instruction rather than the fact.