AI Native · interactive

Streaming UI and partial results

The model writes a JSON answer a few characters at a time. Wait for it all, show the raw text, or parse each partial chunk and render what you can. Press play.

0/24 chunks0 ms

Waiting for the first token…

Received so far

—

At 0 ms: parsed what's arrived so far and closed the open brackets.

Rendering

0 ms

Step 3 of 4 · read left to right

1
Time to first token matters most
Generation takes seconds. Streaming shows progress after a few hundred milliseconds instead.
Perceived speed beats raw speed.
2
Partial output is broken output
Mid-stream JSON doesn't parse, markdown has unclosed bold and code fences, and a word can be cut in half.
{"title":"Reset your rou
3
Parse leniently
Close open strings and brackets, parse, and render the fields you have. Libraries do this for JSON and for markdown.
Keep the layout stable so it doesn't jump.
4
Handle stop, errors and the final check
Let users cancel, show when the stream died, and validate the complete result once it's done.
Partial is for display; validate the final.

In production

Tools and technologies

AI SDKs ship partial-object streaming helpers.

Not tied to one language — use these names when you search docs or design a system.