Team Agents
Early access. Invites, citations and the reports-never-decides rule all work today. Shared administration and per-source ownership are on the roadmap.
When a Team Agent can’t answer, it says so and tells your colleague this one’s for you. The Handover page is where you decide when it does that, and what it says while it does — see Handing over to a person.
What an invite does not include. An invited colleague can ask a Team Agent and read its answers. They do not get an inbox of what other colleagues asked it — a Team Agent keeps no per-person trail, so it never appears under Shared with you in anyone’s twin switcher. On a Support Agent, where the conversations are with your customers rather than your colleagues, an invited colleague can open the Inbox: see who can open a conversation.
A Team Agent is you, for your colleagues. It answers from the internal material you give it — project notes, decisions, how-we-do-things — in your voice, so the fourth person to ask “where are the Meridian call notes?” gets the answer without you.
One rule defines it: it reports, it never decides. Ask it what you wrote and it tells you. Ask it to approve something, commit to a date, or make a call on your behalf, and it says plainly that that’s yours to decide. A colleague acting on a commitment your agent invented is the worst thing that could happen here, so it can’t happen.
Who can read it
In the new app, invite from Workspace → Team → Members.
Only people you invite. An invite is just an email address — no codes, no links: access starts the first time your colleague signs into hiy with that address and opens the agent. Everyone else, including anyone who guesses the URL, gets a 404 — for them it doesn’t exist.
An invite joins your whole organisation, so the same colleague can also open every other agent on this account — your Support Agents, your personal twin and any other Team Agent — and ask any of them anything. What a Team Agent withholds is the trail, not the door: on the others they can also read the conversations recorded there, while nobody on the account can open who asked a Team Agent what.
The address has to match exactly. Membership is keyed to the one verified email you invited — never to a name, a domain, or “close enough”. If your colleague signs in with a different address, they land on a plain “Wrong address” screen naming which invite they followed (with the address masked, so a link that ends up somewhere else doesn’t hand out a full email address) and telling them to sign in with the one that was actually invited. Nothing is granted from that screen; it only tells them what to try instead.
Note
Removing someone removes their access, not just their invitation. It asks first — their address is named in the question, because what they lose is every agent on the account — and if the removal doesn’t go through, the row says so instead of looking done.
Typo’d an address? Each pending invite has its own Change control, so you can fix it in place rather than removing the invite and re-typing it.
hiy emails the invite when you press Invite: the agents they can open, the address to sign in with, the sign-in link, and a Team Agent’s own link to ask it things. That last one matters — a Team Agent never appears under Shared with you, so its address is the one thing a colleague cannot find on their own. The screen says what really happened: Invite sent, or, if email isn’t set up here or the address bounced before, that it didn’t go and the message to send yourself (Copy message, or Copy invite beside their address later). Resend invite, in the ⋯ menu of a pending person, sends the email again. You can send ten invites a minute.
hiy does not build a per-person trail for a Team Agent. Public twins get an attribution view — open the conversation behind a lead or a booking click — and that is deliberately not enabled here. The default for an organisation is that admins see usage and topics, not who asked what, and a transcript view would quietly contradict it.
That survives the account-wide Dashboard. Overview, Insights and Activity add up every agent you own, and a Team Agent’s conversations are left out of every per-conversation figure — it contributes no row to Activity, nothing to the recent-conversations list, nothing to the most-cited tally. It does count in the totals: conversations, questions, people, the activity chart. Top questions counts it too, because a tally of what was asked names nobody who asked it — that is the topics half of the promise above, and it is read separately from anything that could be regrouped into a conversation. All of these are counts about an agent, not a trail about a person, and they are the same numbers a Team Agent’s own dashboard has always shown.
Each block of figures on those pages says which agents it covers, so you can see the exclusion rather than take it on trust — you will read Across 2 of your 3 agents — a Team Agent keeps no per-conversation trail. If a Team Agent is the only agent on your account, Overview and Activity say exactly that rather than showing you an empty list that would read as “nobody asked anything”.
Its entry on the Agents list carries no “last answered” either. Every other agent’s entry shows when it last replied; that is one conversation’s timestamp, and on a three-person team “last answered 3 minutes ago” comes close to saying who just asked. It shows how many questions are waiting to be taught, which is a count.
What colleagues see
Colleagues open the agent at hiy.ai/<handle> after signing in. It greets them by name and shows its starting points, then answers from its sources. The page says AI answers from {Agent}‘s sources · may make mistakes, and it is never listed or indexed.
Invited colleagues see the agents shared with them under Agents › Shared with you, and each one’s Inbox. Anything only the owner can change, such as Plan, Members or API keys, is listed with a lock and says Only the workspace owner can change this.
Turning it on
A Team Agent is a draft until you publish it, and until then the link returns a 404 to your colleagues too. The control is on its Publish tab, under the link, and on Overview. Unpublishing is immediate and acts only on the agent you are looking at — your own public twin and any other agent are untouched.
Publishing hands you the team version of the launch kit: who can reach it, and how an invite works. No QR code and no ready-made posts — they would invite people who get a 404 — and the link itself stays on Publish, beside the list of who may open it.
A new Team Agent starts on Strict scope: it reports what your material actually says and declines the rest, because a confident guess about a company matter is worse than no answer. Change it on the Settings → Rules pill whenever you like.
What it knows
Only what you give it, and what you give it stays internal: sources on a Team Agent can be marked confidential, which makes it impossible — enforced in the database, not by a checkbox — for them to ever reach your public twin.
It shares your monthly allowance
A second agent does not come with a second allowance. Your plan’s visitor messages are one pool for the whole account, and every twin and agent you own draws from it — so adding this one divides the month rather than doubling it. The meter on your Overview and in Settings → Plan & billing already counts it that way. See Plans and billing.
Creating one
A Team Agent is part of Founding, and free during your 14-day trial — making one starts that trial if you haven’t spent it yet. While Early Access spots are open, it starts Early Access instead. Your own public twin is free, always. See Plans & billing.
From the twin switcher in your dashboard: New Team Agent. On the address screen the name field takes your own account name, not the site’s — because a Team Agent speaks as you, and that is exactly the name the identity check compares, so an agent named after yourself goes live without review. (If your account has no display name yet, it’s left empty and you type the name.) Name it after somebody else and a person checks it first; that’s the identity rule doing its job.
If it does go for a check, you’re told once — on the screen straight after you press Create. After that the receipt beside the form keeps showing Name check · pending until you finish, and once you publish the agent, its Overview shows Awaiting review until the check clears. Nothing is lost while it waits: you can import, name and shape it as usual.