Streaming Response

You might say

Show the answer as it comes in, like someone typing, instead of making me wait for the whole thing.

A response sent and displayed in consecutive pieces as it is generatedA chat answer can appear as if it is typing. An early first piece makes waiting feel shorter, but the full answer can still take time to generate and transmit.

Know first

HTTP Token

Streaming

How response pieces become one chat bubble

Illustrative 0.8s · first chunkIllustrative 3.2s · complete

Content will appear here…

Waiting for the first chunk

event 1event 2event 3event 4event 5

How events form a complete response

Pieces arrive: The server sends text or status events; one piece may not be a complete word.

The interface appends them: The product adds each text piece in order to the existing content.

Generation continues: Early content reduces perceived waiting, but total time still depends on the rest.

Quick checkChoose the best answer

A long answer takes 20 seconds, and the product wants earlier visible content while handling failure and completion correctly. How should it work?

Wait for the complete answer, then reveal it slowly with a typewriter animationMark the task complete at the first character and append later content in the backgroundRender a real stream incrementally with generating, complete, interrupted, and retry states

You can say this to an AI Agent

Use the real response stream to update one message incrementally and distinguish connecting, generating, complete, interrupted, and retry states. First content is not completion; preserve partial output on interruption and allow stop.

Learn next

1TTFT / TPS→2Structured Output→3Agent Loop→

Want GPT or Codex?VibeHubGPT support

GPT-6 Astra

Build in 3D. Put your computer to work.