Conversation Panel and Composer
Use the conversation panel to organize discussions, add the right context, choose an execution configuration, and guide work while the Agent is running.
Organize work in the conversation panel
Create a conversation for a new task, or select an earlier conversation to continue its message history. Conversations in the same scope can use the same files, but their message histories remain separate.
Open Summary from the conversation title or the composer to review conversations, tasks, files, and artifacts for the current scope. On desktop, drag the panel’s left edge to adjust its width, or double-click the edge to restore the default width.
Write, attach, reference, and send
Describe the goal, relevant materials, constraints, and expected output. Press Enter to send and Shift+Enter for a new line. When a suggested input appears, press Tab to use it. On the first or last input line, the Up and Down arrows can recall earlier input.
Use the + menu to attach files, search for context, insert a skill, or add an available browser reference. You can also paste files directly into the composer. An attachment is uploaded with the current message; a reference points to existing content.
Review references before sending
References appear as chips above the composer input. Select a chip to open its source, or remove it before sending. You can reference files, artifacts, work items, selected text, and available browser context. Each message can include up to five references.
A reference applies to the request being sent. Pinning or favoriting content helps organize it, but does not make the Agent read it automatically in every conversation. The current preview may appear as a removable reference for the next message.
Choose a model and thinking effort
Recommended choices cover platform automatic routing, faster lightweight work, balanced everyday tasks, and more capable complex work. Available models are grouped by provider; use the descriptions in the menu and the actual choices offered for your site.
Thinking effort controls how much reasoning the Agent applies. Lower levels favor responsiveness; higher levels are useful for difficult analysis and extended tasks. Model and thinking selections are sent with the message and remembered as preferences for later messages.
Add messages while the Agent is working
Steer sends guidance to the active run. Use it to clarify a requirement, correct direction, or add a constraint that should affect the work currently in progress.
Follow up queues a message for a separate later run. Use it for the next request that should wait until the current run finishes. Follow-ups are handled in submission order, and each becomes an independent run in the same conversation.
Choose Steer when the current work should adjust now; choose Follow up when the next task should start afterward. Check the selected mode before sending.