Should I stream responses or wait for the full completion?
New: 16 blocks, from agent runs to voice chat
Build better AI interfaces
Composable React components for chat and agent UIs. Copy them into your project with the shadcn CLI, own the code, ship faster.
npx shadcn@latest add https://ui.spencerwueste.com/r/prompt-input.jsonpandacoderz/ui
How can I help today?
Pick a suggestion or type your own question.
Responses are scripted for this demo. Plug in a transport to go live.
Live block. Responses are scripted; try “Astro islands”, “streaming”, or “tests”.
Components
17 primitives for the AI layer of your app.
Stream it. Tokens appear as they are generated, so the reply feels instant even though total time is the same.
| Approach | First token | Cost |
|---|---|---|
| Streaming | ~200ms | Same |
| Batched | ~4s | Same |
- Read the failing test1.2s
Located the Enter handler assertion.
- Patch the handler
Guarding submit on empty input.
- Re-run the suite
Input
Context window41.2K / 200K(21%)
Requests per minute860 / 1K(86%)
Monthly budget$97.40 / $100(97%)
Requests
4.8k+12%
Tokens (M)
6.4+5%
Spend by model ($)
$165
FableOpusSonnetHaiku
You own the code
Nothing ships as a package. The CLI copies TSX files into your repo and you edit them like any other component.
Built for streaming
Markdown, reasoning, and tool cards all accept partial content and re-render smoothly as tokens arrive.
Agent-readable docs
Every page has a markdown mirror and the site publishes llms.txt, so your coding assistant can read it too.