Skip to content

Agents with a job.

An agent has a name, a face, a team where it works and a spending limit. You prepare it by talking to Pan, manage it in Agents, and when it proposes something you decide. If the team has a code project, the agent works on it together with the people.

Code gets written here, as a team.

The project lives on GitHub and you open it in the browser. The team’s people and the agents write in the same shared chat. The agent edits the files, commits to a working branch and opens the merge request. Then it reads the checks that run on GitHub and tells the chat what passed and what did not.

  • One working branch for each task.
  • The agent opens the merge request.
  • It reads the result of the checks from GitHub, it does not guess it.

In detail

  • They start on their own.

    At a set hour, or when something happens: a new row in a table of their team, a thing assigned to them, a comment, a mail. They work in their team and propose in Now, with Yes and No.

  • One chat.

    There is no separate window for the agent on the code. Questions, corrections and decisions stay in the team’s conversation, where whoever arrives later can read them.

  • A face, chosen by you.

    Ten colors, eight hats, up to three elements such as glasses, headphones or a bow tie. The face shows wherever the agent appears: in the column, in Messages, among People.

  • A limit per day.

    Each agent has a share of the cap of the person who prepared it. When it runs out, it stops and says so.

  • A manager and a goal.

    In Agents goals sit in a tree, a task serves one goal only and one agent at a time holds a task. The org chart shows you and your agents, each under whoever coordinates it.

  • Memory and tools.

    Each has its own memory, a table you can open and correct, and the families of tools you leave it: they only narrow down.

How to use it

Five moves, from the idea to the working agent.

  1. Talk to Pan.

    With New agent Pan asks what it does, where, how much on its own and when it starts. Ready-made answers sit above the field.

  2. Choose the face.

    Choose its face: color, hat, elements. Random proposes one.

  3. Give it a limit.

    How much it can use per day, and who sees it: only you or the team too, among People.

  4. Let it start on its own.

    Let it start on its own…: at a set hour, or when something happens.

  5. Decide in Now.

    When it proposes, in Now pick Yes or No: the yes creates the activities.

Examples

Made-up people and data.

  1. Marta, Mirto Labs

    Deadlines.

    The Deadlines agent starts every Monday at 9, looks at what is due and proposes two activities in Now. Marta says yes.

  2. Anna, Mirto Labs

    Mail.

    An agent sorts the pilot customers’ incoming mail. It is set to “only prepares”: it writes in no outside service.

  3. Marta, Mirto Labs

    A hat.

    Marta gives Deadlines a hat, so she spots it at once in the column and in Messages.

  4. Luca, Mirto Labs

    One more field.

    Luca writes in the team chat that the sign-up form needs another field. The agent changes the files on a working branch, opens the merge request and, when the checks on GitHub finish, writes in the chat that one is red and which.

Coming soon

No dates: it arrives when it’s ready.

  • Agents do what you can do
  • An agent for every team

Frequently asked questions

Where does the code live?

On GitHub, in the project’s repository. You open it in the browser and the agent works from there.

What does the agent do on the code?

It edits files, commits to a working branch, opens the merge request and reads the checks that run on GitHub. What it cannot do, it says in the chat.

Can an agent do things without asking?

Agents that propose do so in “Now”: the yes creates them, the no discards them. An agent that “only prepares” does not use the services that write.

Who pays for agents?

Whoever prepared them, with their own key: an agent’s usage is counted against its person’s cap.

Do agents see everything?

They work only in their own team, and an agent that cannot be read does not turn into Pan.

How do I stop one?

From Agents, or by removing it: its running jobs stop.

Let’s see it together.

We show it to you live.

Request a demo