Reasoning Display

When the AI model uses extended thinking, Hermes-Relay displays the reasoning process as a quiet inline disclosure above the response.

How It Works

The Hermes API Server streams tool.progress events (reasoning/thinking) alongside assistant.delta events (text content). The app collects thinking chunks into a separate block that appears above the main response content.

Appearance

• A live "Thinking…" row with a collapse/expand toggle and no tinted card
• The reasoning text is displayed in a visually distinct style (muted colors, smaller font)
• Open automatically while live and collapsed to "Thought" after completion
• Expandable with a single tap

If you explicitly open or close the disclosure, that choice is preserved when the turn settles. Empty reasoning never creates a row; the separate reply status owns the wait before any reasoning or answer text arrives.

Settings

Toggle reasoning display in Settings > Chat > Show reasoning. When disabled, thinking events are still received but not rendered in the UI.

When Does It Appear?

Reasoning blocks only appear when the model actually uses extended thinking. Not all models or all prompts trigger it. You will see it most often with:

• Complex multi-step problems
• Code generation tasks
• Analytical questions