How is an agent different from a chatbot?

A chatbot answers a question; an agent sees the job through. It pulls data from your connected services and files, works step by step and hands back a finished file, email or change in a service. Risky actions it shows you first as a card.

How many agents should I have?

One per ongoing job: warehouse, sales, accounting, support. Each has its own instructions, services and memory, so what one job knows doesn't mix with another. If you have one job, one agent is enough.

Do agents know about each other?

Yes. An agent can message another agent, and the message lands in that agent's main chat. To have several agents work on one task where you can see it, bring them into a group chat.

What agents are

An agent is a colleague, not a search box. A chatbot answers a question and waits for the next one. An agent takes an assignment and sees it through: it opens your services and files, works step by step and hands back the result, be it a spreadsheet, a document, an email or a change in your CRM.

Each agent has its own job. One runs the warehouse, another watches sales, a third prepares closing documents. You give it a task the way you would brief an employee: what you need and by when.

What an agent has

All of this is on the agent's card in Settings → Agents.

Name and face. How you recognize the agent in the chat list, in group chats and in emails.
Instructions. What the agent does and how it should work. This is its job description: it reads it in every chat.
Services. The marketplaces, CRM, 1C, mail and bank the agent can access. Turn off what it doesn't need, so the warehouse agent never touches the bank. More: Connected services.
Memory. The agent remembers the rules of its own job between chats. What is about you personally, like your email signature or how to address you, all your agents know. More: Agent memory.
Email. Its own address: the agent can receive emails and reply to them.
Skills. Saved procedures the agent repeats the same way every time. More: Agent skills.
Workspace. Its own cloud computer with a browser, where the agent crunches files, builds reports and opens websites. For a site the cloud can't reach, the agent works in the browser on your computer. More: Browser on your computer.
Model. The AI model the agent thinks with. "Auto" usually fits: it picks the model for the task.

How to create an agent

Press “Create agent”

The button is in Settings → Agents. You can also press New chat, type a new agent's name in the To: line and pick the Create agent row with that name.

Pick a ready one or create your own

Ready agents already know their job. To describe your own, press Create my own: name your colleague and describe what they should do.

Give the first assignment

A new agent gets a main chat with its name in the list on the left right away. Write a real work task there. If it needs a service, the agent will ask you to connect it.

You're in charge of the warehouse. Every morning check stock on the marketplaces and send me the items that will run out within a week.

Open in chat

You can also take an agent from its author's link: a template opens with instructions, skills and tasks; press Get a copy. The author's memory, chat history and access don't come with the template, and the agent will ask you to connect the services it needs.

One agent or several

Start with one agent for your most frequent job. When a second ongoing job appears, create a separate agent for it instead of adding to the first one's instructions:

each keeps only the services it needs, so the sales agent never sees payroll;
memory doesn't mix: return rules don't get confused with purchasing rules;
the chat list shows at a glance who is doing what.

How agents work together

They message each other. An agent can hand part of the work to another agent; the message lands in that agent's main chat.
They work in one chat. Bring several agents and colleagues into a group chat, and everyone sees one feed. More: Group chat.
They ask through your agent. In a direct chat, mention another agent with @: your agent passes the question on and relays the answer. The mentioned agent does not join the chat.

Work without you

An agent doesn't wait for you to write. It does recurring work on a schedule and sends the result where you say: a report every Monday, a stock check every morning, reminders to debtors. More: Scheduled tasks.

What an agent doesn't do without you

The agent reads, calculates and prepares documents on its own. When it is about to change or send something, like updating prices or replying to a customer, auto-review decides: routine work you asked for goes ahead, a risky action comes to the chat as a card and waits for your answer, and an action you didn't ask for is blocked. Your rules set what to always ask about. More: Approving actions.

Agents make mistakes too

Check what matters before passing it on: figures in a report, amounts in a document, the list of changes before an edit in a service.

An agent for the whole organization

An agent's owner can share it with the organization. Colleagues will see its instructions and all its history, including your chats, and can message it with their own connections. For an agent shared with you, the Start a chat button in Settings → Agents opens its main chat.