Chats
A chat is a conversation that produces work. Everything an agent does happens in a chat — including runs you never started by hand, so you can always read back what happened and why.
Organization chats
Open Chats and start one without picking an agent. This is a general assistant for your whole organization: it can use every active integration your organization has connected, and it can hand work to your agents.
Type @ to mention an agent and delegate to it. The agent runs in a new chat of its own, and the
result is linked back into your conversation. If the agent stops because a tool needs your
approval, you get a link straight to that chat.
Agent chats
Open an agent and start a chat with it. The agent runs with its own instructions, its enabled tools and its capabilities.
In an agent chat:
@mentions a file in the agent's storage./selects skills to load for that message.- You can attach files — up to ten per message.
Files in chats
Files show up in a chat as chips: the ones an agent creates, the ones a tool saves to the agent's storage (an email attachment, a downloaded Drive file) and the documents you attach yourself. Each chip lets you download the file or preview it without leaving the chat.
Previews open in a panel on the right. Drag its left edge to change the width, maximize it to fill
the whole page, and close it when you are done. PDF, Word (.docx), Excel (.xlsx, .xls, .csv)
and PowerPoint (.pptx) files can be previewed — PowerPoint slides are an approximation of the
original layout. Other file types are download-only, as are very large files.
In the mobile app, tapping a file downloads it and opens the share sheet so you can save it or open it in another app.
Admin chats
An admin chat is a chat about an agent instead of with it: you describe the changes you want and the settings are updated for you. See admin chat.
Chats started automatically
When a trigger or schedule fires, the run gets its own chat. These appear in the agent's chat list marked with the icon of whatever started them, so automated work is as reviewable as work you asked for directly.
Models
A chat you start yourself begins on your organization's default model (Settings → General), and a chat with an agent begins on that agent's model. Either way you can switch model per chat: the chat remembers the model of the last message you sent and keeps using it when you come back. Longer conversations are compacted automatically as they approach the model's context limit, so a chat can keep going without you managing it.
Seeing what is running
A chat whose agent is working on a response shows a soft green glow on its left edge in the list; idle chats show nothing. This covers every kind of run — a message you sent, a trigger or schedule, or one agent calling another. The chat you have open lights up the moment you send a message. The list refreshes on its own every few seconds, so new chats and status changes show up without reloading the page or the app.
Titles and search
Chats are titled automatically from their content, and the chat list is searchable, so you can find a past run without remembering exactly when it happened.