How to build a team of AI agents that work together
One agent is useful. A small team of them, each good at one kind of work, is better: a researcher, a writer, someone who keeps your inbox tidy. On nox.cool you can build that team, give each agent its own job, personality and computer, and let them pass work to each other. Here's how.

Key points
- You can have up to 50 agents on nox.cool, each with its own computer, its own memory and a job written in your words.
- Each agent can have a face, a personality (eleven to pick from, or your own words), a voice and a brain: strongest, balanced or fast.
- Agents hand work to each other, and you can talk to several at once in a group chat by typing their names.
- Projects give a goal to the agents you already have: a lead plans the work, hands out parts and reports back.
- Ready-made agents in the gallery, like a haggler or a morning podcast host, start with their skills and routines already set.
One agent is useful. A small team, each member good at one kind of work, is better. Think of a researcher who digs up facts, a writer who turns them into a clean report, and someone who keeps your inbox under control every morning. On nox.cool you can build exactly that, and let them work together.
Why have more than one AI agent?
You could ask Nox to do everything, and it would. But a team helps in three ways:
- Focus. An agent with one job, written in your words ("Keeps my inbox in order: sorts mail every weekday morning and drafts replies for me to approve"), gets better at that job because its memory fills up with what matters for it.
- Parallel work. Each agent has its own computer, so two agents can browse two different things at the same time without getting in each other's way.
- The right brain for the job. Hard calls deserve the strongest model; sorting receipts doesn't. You can choose per agent.
How do you add an agent?
Tap the plus in the sidebar, give it a name and a job, and pick how it looks and sounds. That's it. Or start from the gallery, where ready-made agents come with their skills and routines already set up. A few examples:
- Haggler reads what you pay for software and bills, finds unused seats and cheaper plans with the evidence, and drafts a counter-offer before each renewal.
- Anchor makes you a short podcast of the news you care about every morning at 7.
You can also just ask. Say "I need help with my freelance invoicing" and Nox can propose a new agent with a name, a job and a first task. It joins your team only when you approve.
What makes each agent different?
Every agent has:
| What you choose | |
|---|---|
| A job | In your own words. It reads this before every task. |
| A computer | Its own, with its own browser, logins and files. |
| A memory | Its own notes, plus a shared layer every agent reads. |
| A face | One of a family of spray-painted smileys that react as it works. |
| A personality | Classic, Fun, Funny, Serious, Calm, Hype, Sarcastic, Coach, Quirky, Sceptic or Butler. Or describe a whole character yourself. |
| A brain | Strongest, balanced (about a fifth cheaper) or fast (cheapest, for simple repeats). |
| A voice | For calls, voice memos and reading replies aloud. |
| Limits | A daily budget and a per-task budget of its own, if you want them. |
How do agents hand work to each other?
When a task fits a teammate better, an agent can pass part of it along. The teammate works on its own computer and reports back into the same task, so you see the whole story in one place.
For example, you ask Nox: "Find five competitors to our product and write a comparison." Nox can ask your Researcher agent to gather the facts with sources, then hand the findings to your Writer agent for the final page. You approve the result, not every step in between.
What are group chats?
Sometimes you want to talk to several agents at once. Type their names at the start of a message:
@Researcher @Writer we need a one-pager on the three best budget laptops for students.
That opens a group chat. Each named agent gets its part straight away and works in parallel. The chat's lead gathers the replies and pulls them together. Their answers show up as they arrive.
How do projects work?
Projects are for goals bigger than one task: launching a small shop, planning a move abroad, getting a newsletter going. You give the project a name, what you want to achieve, any details, and who leads.
The lead (Nox by default) reads the goal, plans the work on one card for you to approve, asks you to connect any apps it needs, and hands out tasks to the agents on the project. The same agent can work on several projects at once. Each project has its own room: a group chat for the team, a scoreboard of what's done, and a short daily review.
Does a team cost more?
Agents cost what their work costs. There's no fee per agent. If you want to keep spending tight, give each agent its own daily and per-task limits, choose the fast brain for simple jobs, and keep an eye on Settings, which shows today's spending on work nobody started, by cause.
The short version
Start with Nox. Add an agent for each kind of work you hand over often, give each a clear job, and let them pass work along. Use group chats for quick team questions and projects for bigger goals. You approve what matters; they handle the rest.
Nox and your team are available today in nox.cool's early access, by invite. Browse the agent gallery, or read what an agent with its own computer can do.
Questions and answers
How many AI agents can I have on nox.cool?
Up to 50. Most people start with Nox and add two or three agents for the jobs they hand over most often.
Do agents share memory?
Partly. Each agent keeps its own notes for its own job, and there's a shared layer every agent reads, for things like your preferences and the people you work with. You can read and edit both.
How do agents work together?
An agent can ask a teammate to do part of a task. The teammate works on its own computer and reports back into the same task. In a group chat you can also address agents directly with @names, and the lead hands out the rest.
Can an agent create other agents?
It can propose one. If you ask for help with a new kind of work, Nox can suggest a new agent with a name, a job and a first task. It joins your team only when you approve.
Does every agent cost more?
Agents cost what their work costs, not a fee each. You can give each agent its own daily and per-task limits, and pick a cheaper, faster brain for simple repeated jobs.
Try it with Nox
nox.cool is in early access. Sign in, or ask for an invite on the home page.
Keep reading

What most AI agents still can't do: five things to check
Most AI agents now share the basics. Five things still set them apart, from what you see before it buys to what a job costs the second time.

What should an AI agent show you before it acts?
An AI agent's approval should show the thing itself: the real page, the price, the email. Why a sentence isn't enough, and how nox.cool built it.

What should an AI agent's screen look like?
An AI agent's app has three jobs: take the task, show the work, ask before anything matters. How we rebuilt every screen of nox.cool around those three.