Agents

Agents list#

Agents in the sidebar lists the Agents you’ve created. Each row shows the Agent’s mark or photo, the latest message from its home conversation, and the time. A dot marks a reply you haven’t read. Pinned Agents stay at the top. A row shows Working… while the Agent is working and Waiting on you when it has asked you a question.

Create an Agent#

1. Click New agent.
2. Enter a name, then describe what the Agent does. You can leave the description blank and add it later.
3. Save the Agent. While Chat drafts its instructions, you may see Writing their brief…. Review the draft before you rely on it.

Creating an Agent doesn’t start a task.

Pin, hide, or delete an Agent#

• To pin an Agent, right-click its row.
• To hide an Agent, Shift+right-click its row, or choose Hide Agent on its profile. The row is removed from the list.
• To delete an Agent’s profile, choose Delete Agent.

Home conversation#

Click a row to open the Agent’s home conversation, a single continuing chat. Tool activity is kept out of the conversation. Charts appear inside the message.

If the Agent needs a connector, the conversation shows a card such as Sign in to followed by the service, with Not now. After you choose, the card shows Signed in, Not signed in, or Not now.

To open the Agent’s profile, tap its name at the top of the conversation.

Appearance#

Under Look:

• Mark — a circle, square, hexagon, diamond, star, or heart, in one of eight colors.
• Upload picture or Replace picture — PNG, JPEG, WebP, or GIF, up to 5 MB.
• Use the mark — removes the photo and shows the mark again.

You upload the picture yourself. If you ask, the Agent can change its mark’s shape or color, or remove its photo. Agents don’t generate their own portraits.

Instructions and access#

Instructions tell the Agent how to write and what to do in every conversation. Instructions can be up to 8,000 characters. Click Save after you edit them.

Code sets limits for Code runs that use this Agent. Files can be Read-only or Project write. Memory can follow your account setting, be read-only, or be off. The current task can narrow these limits but can’t widen them.

Skills and Connectors are switches. When one is on, this Agent can use it. Turning one off doesn’t affect the others. A skill must be installed in your library before an Agent can use it. See Skills and plugins.

Routines#

Routines lists the Agent’s schedules, with states such as Active, Running, Paused, Done, Cancelled, and Needs attention. Add opens the home conversation with a draft that starts with Set up a routine to. Nothing is scheduled until you ask. Scheduled tasks explains where routines post and why their model can’t be changed.

Use an Agent in a chat or in Code#

In a new chat, choose the Agent in Advanced before you send the first message. The Agent then stays with that chat. None means no Agent.

In Code, ask for the Agent by name and say what it may change. For example: “Ask the review agent to inspect these changes without editing files.” Whether Code can hand off the work depends on the task’s permission and delegation mode. The Agent’s home conversation and the Code transcript are separate. See Agents in Code.

Dynamic Agents in account settings is separate from the Agents you create and from delegation mode in Code, which controls whether a Code task can hand off work.

Open an Agent’s computer from iPhone#

In the Agent’s home conversation on iPhone, tap Open this Agent's computer, then choose a paired computer. AI Integrator for iPhone connects to it and opens a new browser tab there for the Agent. If you haven’t paired a computer, the list says so. The desktop app must be running on the computer. Access to windows and screens depends on the Mobile Gateway permissions described in Remote access.