Thinking process

Some models expose their thinking process — the intermediate reasoning the model produced before its final answer. Pegasus surfaces this as a collapsible "Thinking process" block inside the message.

Expanded thinking process block

When it appears#

The thinking block appears whenever the model used produces thinking content. Not every model emits it. You'll typically see it on advanced models when asked complex questions.

Toggling visibility#

Click Show thought process or Hide thought process to expand or collapse the block. The state is per-message, not global.

A duration ("12s") is shown next to the toggle — how long the model spent on the thinking phase.

Why it's useful#

• Debugging answers — see why the bot reached a conclusion. If the thinking is wrong, you can correct the question or refine the bot's documents.
• Trust — verifying that the bot is reasoning from your documents and not hallucinating.
• Learning — sometimes the thinking is the interesting part (multi-step reasoning, comparing sources).

Limitations#

• Thinking is shown only after the final answer arrives, not streamed in parallel.
• The thinking content does not count separately for credit — it's part of the same response.
• The thinking is not preserved when you copy the response (only the visible answer is copied).