Reasoning

Collapsible "thinking" block for AI reasoning/thought tokens. ReasoningTrigger shows a shimmering "Thinking..." while isStreaming, then "Thought for N seconds" (duration measured automatically or passed in); ReasoningContent renders the reasoning text as markdown with Streamdown. Auto-opens when streaming starts and auto-closes one second after it ends; defaultOpen={false} disables the auto-open, a reader's manual toggle is respected, and a controlled parent receives both requests through onOpenChange. Use it above a message response when the model exposes reasoning.

pnpm dlx shadcn@latest add @uifiles/reasoning

Registry JSON Source

Installs with

Streaming

Opens itself while the thought streams in and closes one second after it ends.

Finished, open

A duration was supplied, so the trigger reads how long the model thought. Open by default; the fence, table and formula scroll in their own boxes.

The user wants a caching strategy for a profile page.

Profile data changes rarely, so a short revalidation window is fine.
no-store would hit the origin on every request.
force-cache would serve stale names after a rename.

Recommend revalidate: 60 and mention revalidateTag for instant invalidation on edit.

Checking the route config before answering:

Caching strategy	Freshness guarantee	Origin load	Recommended for
`no-store`	Always fresh	Every request	Personalized dashboards
`revalidate: 60`	Up to 60s stale	Once a minute	Profile pages
`force-cache`	Until redeploy	Once per build	Reference data

If a model predicted which profiles get renamed, its regularized logistic loss would be:

\mathcal{L}(\theta) = -\frac{1}{n} \sum_{i=1}^{n} \left[ y_i \log \sigma(\theta^\top x_i) + (1 - y_i) \log\left(1 - \sigma(\theta^\top x_i)\right) \right] + \lambda \lVert \theta \rVert_2^2

Not worth it for a profile page: a 60 second window is simpler.

Finished, collapsed

No duration, so the trigger only says the model thought. Collapsed until opened.