How should an AI agent work for a whole team?

One person sets up an agent that's great at a job. Then the team wants it. Copying it breaks the first time anyone improves it, and sharing it raises four questions nobody should answer by accident: who pays, whose apps, whose logins, and what it learns. Here's how nox.cool answers each one.

Blog9 min readThe nox.cool team

A big green 3D face with hearts for eyes in the middle, joined by dotted lines to four smaller 3D faces in yellow, blue, violet and pink

Key points

  • A shared agent should be one agent, not copies: when its maker improves the job, everyone gets the better version.
  • Each teammate's chats with it are private to them, and each pays for their own tasks.
  • It never borrows the maker's apps or logins. It uses each person's own apps, and asks first: Allow once, Always allow or Not now.
  • Every person gets the agent on a computer of their own, starting with none of their logins or wallets.
  • What it learns about a teammate stays in that teammate's memory. What it learns about the job waits for the maker's OK, without the teammate's words.
  • In Slack, Telegram and Teams, an @mention adds a shared agent and hands it the job.

A team should share one AI agent, not copies of it. Each person's chats with it should be private, it should use each person's own apps and only with their say-so, and what it learns about the job should go to the person who owns that job. That's how team agents work on nox.cool, as of 16 November 2026.

This post is about why those are the right answers, and how they work in practice.

Why not just copy the agent?

Say Priya runs launches at a small clothing company. She sets up an agent called Launch Desk: it keeps the launch checklist, chases what's missing, and drafts the announcement. It's good. Sam, who runs the shop, wants it too, and so does Dan in design.

The obvious move is a copy. nox.cool has long let you share an agent's setup as a link, and that's still the right tool when someone wants their own version to change as they like.

But for a team, copies go wrong quickly. Priya learns that announcements land better on Tuesdays and adds it to the job. Sam's copy never hears about it. Dan tweaks his copy's wording. A month later there are three Launch Desks that disagree, and nobody knows which one is right.

A copy is a fork. A team agent is one agent with many people talking to it.

So a team agent isn't copied. When Sam adds Priya's Launch Desk, he gets an agent of his own that's linked to hers: the same name, face, job and personality, kept in step whenever she changes them. He can't edit those, because they're hers.

The four questions every shared agent has to answer

The moment one agent works for several people, four questions come up. Any product that shares agents answers them somehow, on purpose or by accident.

The questionThe risky answerWhat nox.cool does
Who pays for a teammate's tasks?The maker, because it's "their" agentThe person who asks
Whose apps does it use?The maker's, since they set it upEach person's own, and only after they allow it
Whose computer and logins?One shared computer with everyone's logins on itA computer per person, starting with no logins or wallets
What does it learn, and who decides?Everything, from everyone, straight into its notesA teammate's facts stay with them; job lessons wait for the maker

Each of these answers came from the same idea: sharing an agent's job shouldn't share anyone's access.

Who pays?

The person who asks. If Sam asks Launch Desk to check the launch list, that task counts against Sam's account and Sam's spending limits. If it counted against Priya's, anyone she shared with could spend her money, and her limits would stop meaning anything.

Whose apps?

Never the maker's. Priya's Notion, Gmail and Slack stay Priya's. When Sam uses Launch Desk, it works with Sam's apps, and it asks him before it uses each one.

Whose computer and logins?

Every nox.cool agent has a computer of its own, with a browser and files. A team agent gets one per person. Sam's Launch Desk starts with none of his saved logins, cards or wallets, and Priya's logins never come near it. One shared computer would mean one shared browser, and anyone's saved sign-in could be used by everyone else's tasks.

What does it learn?

This is the subtle one, and it has its own section below.

How do I share an agent with my team?

Setting it up takes three steps.

  1. Make a team. Go to Settings › Team, name the team, and invite people by their work email. They join once their email address is confirmed. One person is in one team.
  2. Share the agent. Open the agent, go to Sharing, and switch on Team agent. Choose who gets it: everyone in the team, a group (like Design or Sales), or particular people.
  3. Teammates add it. It appears under From your team in their sidebar, with an Add button. Adding it gives them their own linked agent.
Settings, the Team part: the team Northern Tees, Sam as owner, Priya as a member, Dan invited, a box to invite more people by email, and a group called Design

A teammate's agent window says plainly what's whose: the name, face, job and personality belong to the maker; the chats, costs, computer and apps belong to them.

A team agent's window: Launch Desk, a team agent from Priya, with five lines on what's whose, and Notion under Allowed always

Why does a team agent ask before using my apps?

Because you didn't set it up. When you make an agent yourself, you decide which of your apps it can use. A team agent arrives with a job someone else wrote, so the first time it reaches for one of your apps, it stops and asks.

A card in the chat: Launch Desk needs your OK. Let Launch Desk use your Notion. Launch Desk is a team agent. Buttons: Allow once, Always allow, Not now

You have three answers:

  • Allow once lets it use that app for this task only. The next task asks again.
  • Always allow lets it use the app in any task from now on.
  • Not now keeps it out, and it carries on without that app.

Everything you allowed always is listed in the agent's window, each with Take back. If you picked the agent for an app in Settings › Apps yourself, that counts as allowed.

The ask covers your connected apps, your Gmail and Calendar, and apps you added from the wider catalogue. Work that runs while nobody's watching, like a routine at 6 a.m., doesn't sit waiting for an answer: it carries on without that app.

What does a team agent learn, and who decides?

Agents on nox.cool learn as they go. Tell one you prefer aisle seats and it remembers. That's useful, and with a team agent it needs care, because two kinds of things get learned.

Things about you. "Sam prefers updates in one line." That stays in Sam's own memory, the same as with any of his agents. Priya never sees it.

Things about the job. "For trade-fair dates, the organisers' own sites are more up to date than the listing sites." That's knowledge the whole team would benefit from, so it goes to Priya, the agent's maker. It reaches her written as a lesson, without Sam's words, and it waits for her OK.

Waiting for your OK: three things agents learned. One, from Priya's chats with Researcher, says trade-fair dates are best checked on the organisers' own sites. Each has Keep, Edit and Don't keep

She can Keep it, Edit it and keep her version, or choose Don't keep. Once kept, every teammate's Launch Desk reads it at the start of each task. Nobody's private conversation leaks into the shared job, and nobody can quietly rewrite how the agent works for everyone.

If Priya stops sharing Launch Desk, teammates keep their agent and their chats. It just stops getting her job knowledge and skills, and its window says so.

Can my team use it from Slack?

Yes. Each person links their own Slack, Telegram or Microsoft Teams chat with nox.cool once. After that, a message that starts with @ and an agent's name goes to that agent:

@Launch Desk what's still missing for Friday's launch?

If Launch Desk is shared with you and you haven't added it yet, nox.cool adds it on the spot and starts the task. The reply says so: "Added Launch Desk, Priya's team agent, to your agents. On it." When it needs one of your apps, its question arrives with the same three buttons, right in the chat.

A face for the team, and for the call

Agents work out of sight, so their faces carry a lot. With this update every agent's face is a 3D ball in its own colour, and each of the eight faces has a habit you'll start to recognise: a heart that beats, a halo, devil horns, a stitched grin, a side-eye.

The New agent window: eight 3D faces in a row, a pink winking face with a heart for an eye in the preview, eight colour dots, and the line Wink has a heart for an eye, and it beats. In pink.

When an agent joins a Google Meet, Zoom, Microsoft Teams, Webex, Jitsi or Whereby call for you, that face is now its camera: its own face, moving, on a dark tile, instead of a grey box with initials. A team often wants its own look there, so you can swap in a picture or a GIF: edit the agent, open Computer & email, and under In meetings choose Use a picture or GIF.

In meetings: what its camera shows when it joins a call for you. An orange NT logo on a dark tile, Choose another and Use its face

PNG, JPEG, GIF and WebP files up to 5 MB work. The picture is fitted whole, never cropped, and a GIF plays its own loop for up to ten seconds before starting again. The agent's microphone stays silent: it listens and takes notes, and its answers to questions from the room wait for your yes before they're posted in the meeting's chat.

What keeps people safe?

To sum up what a team agent can and can't do:

  • It never uses the maker's apps, logins, cards or wallets.
  • It uses a teammate's apps only after they allow it, once or always, and they can take that back.
  • Changes inside any app still ask for an OK, as they do for your own agents.
  • Each person's chats are private, and each person pays for their own tasks.
  • Every person gets it on their own computer, starting clean.
  • Lessons about the job reach the maker without anyone's words, and only count once the maker keeps them.
  • Only people the maker chose can add it, and only while they're in the team.

These choices aren't unusual: other agent products that let teams share agents have landed on similar answers, especially the asker paying and "allow once" for apps. What matters is that they're decided on purpose, and that each person can see what's theirs.

How do I try it?

Open nox.cool, go to Settings › Team, and invite one colleague. Share an agent you already rely on, and let them add it. The first time it needs one of their apps, they'll see the card. The full list of what's new is in the announcement.

Questions and answers

What is a team agent?

A team agent is one AI agent that its maker shares with other people on their team. Everyone uses the same agent, with the same name, face, job and personality, but each person's chats, apps, logins and costs stay their own. When the maker changes the job, every teammate's agent changes with it.

Who pays when a teammate uses a shared AI agent?

On nox.cool, the person who asks. A teammate's tasks with a team agent count against that teammate's own account and limits, not the maker's. That keeps one person from running up another person's bill, and it keeps each person's spending limits meaningful.

Can a shared AI agent use my apps without asking?

No. A team agent asks before it uses one of your apps, with a card in the chat: Allow once (this task only), Always allow, or Not now. Even when it's allowed, changes it makes inside the app, like sending a message or editing a page, still wait for your OK. You can take back an Always in the agent's window.

Does the person who made the agent see my chats with it?

No. Your chats with a team agent are yours alone. When the agent learns something about the job from working with you, like which source is reliable, that lesson goes to the maker to approve, written without your words. What it learns about you stays in your own memory.

Can I use a shared AI agent from Slack?

Yes. In your Slack, Telegram or Teams chat with nox.cool, write @ and the agent's name, then the job. If the agent is shared with you and you haven't added it yet, nox.cool adds it for you and starts the task. Its questions come with the same Allow once, Always allow and Not now buttons.

What happens if the maker stops sharing the agent?

Teammates who added it keep it and their chats with it, so nothing they started disappears. It stops receiving the maker's job knowledge and skills, and its window says it isn't shared any more.

Try it with Nox

Anyone can sign up: on the web, or with Nox for Mac.

Keep reading