Duties, the work agents do on their own.

Expand

How a duty starts

One duty can start in several of these ways, as long as each hands the agent the same kind of thing: nothing to work on (a schedule and the Run button), one item at a time (buttons on items, a schedule for each item, and changes to items of that kind), a message from outside, or a conversation. A button and a new email never share a duty, because the email brings what the duty works on and a button brings the item it was pressed on: make two duties instead. The set-up greys a way that cannot join the ones you chose, and says why. A duty has at most 10 ways to start: one Run button, and up to 5 each of schedules, buttons on items and schedules for each item.

Schedules can repeat every few minutes or hours, daily, monthly, once, or on a custom pattern; the set-up shows the next three times it will run. How often a duty can run depends on your plan. Changes an agent makes itself never start its own duties.

Give an agent a duty

Open the agent's Duties tab and press Add a duty. You need to be able to manage the agent. The set-up opens over the page in five short pages, and the agent asks each question itself.

1. 1

   How it starts

   Choose one or more ways, then the details of each: when a schedule runs, where a button shows, and which changes or messages count. The time zone its schedules follow is on this page too.
2. 2

   What it does

   Name it and write its instructions in plain words (up to 8,000 characters), or start from an example.
3. 3

   What it may use

   A duty can only narrow what its agent can do, never widen it: the places it may read, its tools (only what this duty needs, or all the agent's tools), the connected mailboxes it may use, each at a level (read only, read and draft, read, draft and send, or send only), and web pages.
4. 4

   How far it may go alone

   Outside Stable Baseline: drafts only, ask before each send, or send on its own. Inside Stable Baseline it can be stricter than the agent's usual approvals, never looser. Name an owner and up to five backups to approve its requests, which wait up to 72 hours, and set limits if you like: runs a day and a monthly credit limit.
5. 5

   Review

   The duty's card as it will look, and each page in a line you can change with one press. Then Save duty.

Replies to mail

• When to reply: only when a message needs it, or every message.
• Who replies go to: everyone on the message, the sender only, or the people you list.
• Allowed email domains: set by owners and admins. A duty never replies to anyone outside them.
• A signature.

Only owners and admins can let a duty send mail on its own.

Buttons on items

A button on items puts the duty in the Agents menu of the items you choose: documents, whiteboards, plans, tasks, improvements, database rows or dashboards. Choose where it shows: everywhere, in a workspace or a project, in a folder (for documents and whiteboards), or on up to 50 chosen items; a button on rows shows on the rows of one database or one of its tables. On plans, tasks and improvements it can also wait for conditions, such as a status, a priority, being late or, on tasks and improvements, being yours. Or describe where it should show in your own words, and the set-up reads them into the same choices for you to check.

Expand

A button shows only to people the agent is shared with, on items they can open. Pressing it runs the duty on that item for the person who pressed it: the agent does only what the agent, the duty and that person all allow, and anything it must ask first is asked of that person. The menu stays open while the agent works, and says how the run ended.

A duty that is handed nothing to work on can have a Run button instead, on its agent's card and in the agents panel in the top bar. It runs the duty now, for you, the same way.

Duty cards

An agent's Duties tab shows each duty as a card: its state (when it runs next, working, listening or paused) and three groups of pictures, how it starts, what it uses and what it makes. Rest the pointer on a picture, or tap it, to read it in words. The bottom of the card says how many of the three risky things it does, how much it ran, and holds its Run button when it has one.

Expand

Watch it work

Each duty has its own page: what it does in one sentence, the same pictures, its health, today's runs, recent runs, credits and setup. A run is waiting to start, running, waiting for approval, paused, done, failed or cancelled, and it says how it started: on its schedule, when something changed or an email arrived, or who pressed its button. When a run needs a person, its proposal appears in the agent's Approvals and on the Approvals page.

Expand

Expand

What a duty costs

A run someone starts with one of the duty's buttons costs the same as any other run. Filters are checked before any work starts, so events that do not match cost nothing. Asking an agent yourself stays free.

• Set a monthly credit limit on one duty, or Set a cap for all of an agent's duties. Owners and managers are told when an agent reaches 80 per cent of its cap, and at a limit the duty pauses until next month.
• If your organisation runs out of credits, duties that use them pause and start again by themselves once there are credits. What arrives meanwhile is kept for up to 7 days.

Pause a duty

Pause a duty and what arrives is kept, and runs when you resume it. Owners and admins can Pause every duty in the organisation at once; what arrives is kept for up to 7 days. A paused duty's buttons leave the menus until it is resumed.

Who can do what

• Anyone who manages an agent can add duties to it.
• A duty's owner can also edit or delete it, and its backups can pause it.
• Anyone the agent is shared with can press its duties' buttons, where they show.
• Limits on duties and runs depend on your plan. Usage, at the top of the Agents page, shows them. See pricing.

See Security and access for whose access a duty works with, and who may approve its work.