The Chat Input is the editor at the bottom of every agent card. It’s where you type prompts, attach context, and control how the agent behaves.
Basics
- Enter — send message
- Shift+Enter — new line
- If you send a new message while the agent is running, your message gets queued and dispatched automatically when the agent finishes.
- You can queue multiple messages.
- Queued messages can be edited, deleted, or re-ordered prior to them getting sent to the agent.
/ Slash Commands
Type / within the chat input to open the command picker:
- Templates — insert a reusable prompt template (with optional fill-in fields)
- Skills — manually attach a skill, forcing an agent to use it.
- Modes — switch the agent’s mode (same as manually selecting the mode in the chat input via the dropdown)
@ Context Commands
Type @ at the start of a word to open the context picker:
@file — browse and attach a local file or folder as context@web — force the agent to use WebSearch/WebFetch for this message- Integrations — force specific tools (e.g. Google Workspace, X/Twitter, Reddit).
Attachments
Images & Files
Drag and drop, paste from clipboard, or use the attachment button. Thumbnails appear above the input.
Element Selection
Click the select icon in the input toolbar to enter element selection mode. This lets you click anything on the canvas to attach it as context:
- Other Agents — allows the current agent to access the context of a prior agent and re-deploy a copy of this prior agent as it’s own employee.
- Browsers — allows the current agent to access the context of and control a browser.
- Apps — allows the current agent to access the context of and control a browser.
Selected elements appear as chips above the input.
If you copy anything in a workspace with ⌘C and paste into the input with ⌘V, they’re converted into selection context automatically.
Mode & Model Selectors
Two pill buttons at the bottom of the input let you switch the mode and model for this session. See Agents overview for the available options.
Context Window
A small ring next to the selectors shows context usage. Hover for the exact count (e.g. 45.2% · 90.4K / 200K).
Stopping / Resuming Agents
While an agent is running, the send button in the chat input turns into a stop button. When clicked, this stops an agent and of it’s any sub-agents. If an agent has been stopped, you will also see a resume button in the chat history that enables you to let an agent continue where it left off. 
