Skip to content
GetAPI
English
English简体中文繁體中文日本語한국어EspañolPortuguês (Brasil)FrançaisDeutschРусскийItalianoالعربية
Menu

GetAPI.ONE

Stream Responses from GetAPI.ONE

Consume a server-sent event stream, append only text deltas, and treat completed, failed, and incomplete events as distinct terminal outcomes.

Choose a compatible path

  • Use a current catalog model whose declared endpoint is Responses.
  • Keep the API key in a trusted server or local process.
  • Decide how the caller will display partial text and recover from an interrupted stream.
text
POST https://www.getapi.one/v1/responses
Accept: text/event-stream
Authorization: Bearer $GETAPI_ONE_API_KEY

Read the event lifecycle

  1. Send stream=true through a server-side SDK or HTTP client with automatic retries disabled for the first controlled request.
  2. Append only response.output_text.delta payloads to visible text.
  3. Stop on response.completed and keep the final response metadata needed by your application.
  4. Surface response.failed and response.incomplete as non-success outcomes; do not label partial text complete.
  5. Treat an error event, transport exception, EOF before completion, or local cancellation as non-success and close the stream.
python
import os
from openai import OpenAI

terminal_event = None
try:
    with OpenAI(
        api_key=os.environ["GETAPI_ONE_API_KEY"],
        base_url="https://www.getapi.one/v1",
        timeout=30.0,
        max_retries=0,
    ) as client:
        with client.responses.create(
            model="<RESPONSES_MODEL_ID_FROM_CURRENT_CATALOG>",
            input="Give three concise deployment checks.",
            stream=True,
        ) as stream:
            for event in stream:
                if event.type == "response.output_text.delta":
                    print(event.delta, end="", flush=True)
                elif event.type == "response.completed":
                    terminal_event = event.type
                    print("\ncompleted")
                    break
                elif event.type in {"response.failed", "response.incomplete", "error"}:
                    terminal_event = event.type
                    raise RuntimeError(f"stream ended with {event.type}: {event}")
except KeyboardInterrupt:
    raise SystemExit("cancelled locally; the stream and client were closed")

if terminal_event != "response.completed":
    raise RuntimeError("stream ended before response.completed")

Map events to UI state

EventApplication action
response.output_text.deltaAppend event.delta; do not replace previously rendered text.
response.completedMark success and close loading state.
response.failedShow the structured error and a safe retry action.
response.incompleteKeep partial output visibly incomplete and inspect incomplete details.

Verify the expected result

  • Text appears incrementally without duplicated deltas.
  • Loading ends only after a terminal event or transport error.
  • An interrupted connection never displays a false success state.

Handle transport and protocol errors

  • If no events arrive, verify the exact /v1/responses route, stream=true, proxy buffering, and the model endpoint declaration.
  • If output repeats, append each delta once and do not combine SDK accumulated output with raw events.
  • Retry only before visible side effects or with application-level idempotency; a disconnect does not prove the upstream request did nothing.

Keep streaming server-side

text
Browser → your authenticated backend → GetAPI.ONE
                 server-held key only

Next steps