Reasoning and context

Reasoning controls what a compatible model generates and how you read its work. Context describes how much information the model can use for a response.

Two different reasoning controls

Control	Where to find it	What changes
Reasoning	Context indicator beside the composer → Generation	Whether reasoning is requested for new responses from compatible models.
Stream Reasoning	Settings → App	Whether reasoning is displayed as it arrives.

When Stream Reasoning is off, the app shows a compact Thinking indicator while the model reasons. Completed reasoning remains available in the assistant message’s collapsible section. This display setting does not turn reasoning generation off.

The generation preference is shared by the app; it is not a separate saved value for each conversation. Its effect depends on the selected model’s support.

Read the context indicator

Open Context and model details beside the composer. The sheet can show:

• Capabilities: what the selected model supports.
• Context: Used, Remaining, Context Length, and Model Maximum when available.
• Latest Response: input, output, and reasoning token counts returned by the server.
• Processing & Budget: relevant runtime details reported for the model.

These details are measurements and model information, not controls for changing the host’s context allocation. Missing statistics do not mean zero usage; some values appear only after a response or when the server reports them.

Loaded context versus model maximum

A model can be loaded with less context than its theoretical maximum. The active context length determines the current working limit. Larger context allocations can require more host memory.

Follow a response

With Auto-scroll Responses disabled in Settings → App, a new assistant response starts near the top and streams in place. Enable it if you prefer following the reply as it grows. You can scroll manually and use Scroll to latest message to return to the end.

Reasoning and tool activity can make a response take longer than a short text answer. Use Stop to cancel if needed. For an interrupted response or a context-limit error, see Models and responses.