The composer
The composer
The composer is the text input area at the bottom of the chat screen. It is where you write prompts, attach files, trigger slash commands, and control model settings before sending a message.
[SCREENSHOT: Composer bar showing text area, model picker, thinking-level button, attachment button, and send button]
What you see
The composer bar contains:
- Text area — grows vertically as you type; supports plain text and multi-line input.
- Model picker — shows the currently active model. Click to switch models for the session.
- Thinking level — a toggle that cycles through
off,low,medium, andhighreasoning effort. On Claude 4.6 the default isadaptive. - Attachment button — opens the file picker to attach images or documents.
- Voice button — records a voice message and transcribes it into the text area.
- Send button — submits the message. Disabled while a response is streaming; replaced by a Stop button you can click to abort.
- Web search toggle — enables web search augmentation when the gateway supports it.
Sending a message
Type your prompt and press Enter to send. To insert a real newline without sending, press Shift+Enter.
On desktop you can also use Cmd+Enter (Mac) or Ctrl+Enter (Windows/Linux) to send without moving your hands from the keyboard.
Switching models mid-session
Click the model name in the composer to open the model picker. Select a different model; the change takes effect on the next message. The chosen model is written to the gateway config so it persists across page reloads. Local models (Ollama) are tracked client-side only and are not written to the gateway.
To open the model selector with the keyboard, press Cmd+Shift+M (Mac) or Ctrl+Shift+M (Windows/Linux).
Adjusting thinking level
Click the thinking-level icon to cycle through the available effort tiers. Higher levels give more thorough reasoning at the cost of longer response time. Set it to off to disable extended thinking entirely.
Attaching files
Click the paperclip icon or drag a file onto the composer to attach it. See Attaching files for supported types and size limits.
Using slash commands
Type / at the start of a message to open the slash command menu. See Slash commands for the full list.
Interactive clarification (mid-turn)
When an agent invokes the clarify tool mid-turn, the chat area inserts a Clarification needed card above the composer. The card shows the agent’s question and, when the agent supplies a list of choices, a button for each option.
You can:
- Click a choice button to submit that answer immediately.
- Type a free-text reply in the input field and press Send (or Enter).
Submitting an answer POSTs to /api/sessions/:key/clarify with your reply, which unblocks the agent’s paused turn on the gateway. The card updates to “Answer submitted — resuming…” while the turn continues. If an error occurs the card shows the error inline and lets you retry.
The clarify card requires gateway support. Sessions using the OpenAI-compatible fallback path do not support mid-turn clarification.
Aborting a response
While the model is generating, the Send button becomes a Stop button. Click it to abort the current stream. Partial output already in the thread remains visible.
Common issues
Enter sends when I want a newline — Use Shift+Enter for newlines. There is no setting to swap Enter and Shift+Enter.
Model picker shows no models — The gateway may not be running, or no provider is configured. See Connecting your AI provider.
Voice button is greyed out — Browser microphone permission may have been denied. Check your browser site settings and reload.