Streaming

Every run adapter supports streaming. Pass stream: true and iterate the returned AsyncIterable. Events are always shaped as the OpenAI SDK types — Supercompat translates provider-native deltas into OpenAI events on the way back.

Responses API

const stream = await client.responses.create({ model: 'gpt-4.1-mini', input: 'Count to three.', stream: true, }) for await (const event of stream) { if (event.type === 'response.output_text.delta') { process.stdout.write(event.delta) } }

Event types you'll see (OpenAI.Responses.ResponseStreamEvent):

response.created

response.in_progress

response.output_item.added / response.output_item.done

response.output_text.delta / response.output_text.done

response.function_call_arguments.delta / response.function_call_arguments.done

response.completed

All providers emit this vocabulary.

Assistants API

const stream = await client.beta.threads.runs.create(thread.id, { assistant_id: assistant.id, stream: true, }) for await (const event of stream) { switch (event.event) { case 'thread.message.delta': { const delta = event.data.delta.content?.[0] if (delta?.type === 'text') process.stdout.write(delta.text.value) break } case 'thread.run.requires_action': // Handle tool calls — see Tools doc. break } }

Events follow OpenAI.Beta.AssistantStreamEvent:

thread.created

thread.run.created / thread.run.in_progress / thread.run.requires_action / thread.run.completed / thread.run.failed

thread.run.step.created / thread.run.step.delta / thread.run.step.completed

thread.message.created / thread.message.delta / thread.message.completed

Translated under the hood

With completionsRunAdapter, Supercompat streams the provider's /chat/completions response and re-emits each chunk as the matching OpenAI event. With geminiRunAdapter or openaiResponsesRunAdapter, deltas come from the provider's native streaming endpoint and are translated on the fly.

Your iteration code stays the same whether the backend is OpenAI, Anthropic, Gemini, or a Completions-compatible provider.

Edge runtimes

If you're running on an edge platform (Vercel Edge, Cloudflare Workers), pass waitUntil to the OpenAI or Azure Responses run adapter so background work — such as writing the final response to storage — is finished even after the stream is returned to the caller.

import { openaiResponsesRunAdapter } from 'supercompat/openai' supercompat({ clientAdapter, storageAdapter, runAdapter: openaiResponsesRunAdapter({ waitUntil: (p) => context.waitUntil(p), }), })