Stream tokens and tool deltas with unified streaming ergonomics across all providers.

Real-Time Response Streaming

Every provider in raseo-sdk implements .stream(request) returning a unified ModelStreamResponse.

`ModelStreamResponse` Interface

export interface ModelStreamResponse {
  // 1. High-level text chunk stream
  readonly textStream: AsyncIterable<string>;

  // 2. Typed event chunk stream
  readonly stream: AsyncIterable<ModelStreamChunk>;

  // 3. Accumulated final response promise
  readonly response: Promise<ModelResponse>;
}

1. High-Level Text Streaming (`textStream`)

The simplest and most common way to stream assistant tokens:

import { OpenAIProvider } from "raseo-sdk/openai";

const provider = new OpenAIProvider({
  apiKey: process.env.OPENAI_API_KEY!,
  model: "gpt-4o-mini",
});

const streamResult = await provider.stream({
  messages: [{ role: "user", content: "Write a short poem about the ocean." }],
});

for await (const text of streamResult.textStream) {
  process.stdout.write(text);
}

// Await accumulated response metadata
const final = await streamResult.response;
console.log("\nFinished because:", final.finishReason);
console.log("Tokens used:", final.usage?.totalTokens);

2. Typed Delta Events (`stream`)

If your UI needs to react to typed event chunks (e.g. distinguishing text chunks from tool call deltas):

for await (const chunk of streamResult.stream) {
  switch (chunk.type) {
    case "text-delta":
      console.log("Text delta:", chunk.text);
      break;

    case "tool-call-delta":
      console.log("Tool call delta:", chunk.toolCall);
      break;

    case "finish":
      console.log("Stream completed. Reason:", chunk.finishReason);
      if (chunk.usage) {
        console.log("Usage:", chunk.usage);
      }
      break;
  }
}

Zero Provider Differences

Whether you invoke OpenAIProvider, AnthropicProvider, or GeminiProvider, the streaming interface and guarantees remain identical:

Same textStream async iteration.
Same typed event shapes.
Same final .response promise resolving token usage metrics.

Real-Time Response Streaming

Real-Time Response Streaming

ModelStreamResponse Interface

1. High-Level Text Streaming (textStream)

2. Typed Delta Events (stream)

Zero Provider Differences

On this page

`ModelStreamResponse` Interface

1. High-Level Text Streaming (`textStream`)

2. Typed Delta Events (`stream`)