`createAgentUIStream`

The createAgentUIStream function executes an Agent, consumes an array of UI messages, and streams the agent's output as UI message chunks via an async iterable. This enables real-time, incremental rendering of AI assistant output with full access to tool use, intermediate reasoning, and interactive UI features in your own runtime—perfect for building chat APIs, dashboards, or bots powered by agents.

Import

import { createAgentUIStream } from "ai"

Usage

import { ToolLoopAgent, createAgentUIStream } from 'ai';

const agent = new ToolLoopAgent({
  model: "anthropic/claude-sonnet-4.5",
  instructions: 'You are a helpful assistant.',
  tools: { weather: weatherTool, calculator: calculatorTool },
});

export async function* streamAgent(
  uiMessages: unknown[],
  abortSignal?: AbortSignal,
) {
  const stream = await createAgentUIStream({
    agent,
    uiMessages,
    abortSignal,
    // ...other options (see below)
  });

  for await (const chunk of stream) {
    yield chunk; // Each chunk is a UI message output from the agent.
  }
}

Parameters

agent:

Agent

uiMessages:

unknown[]

abortSignal:

AbortSignal

timeout:

number | { totalMs?: number }

options:

CALL_OPTIONS

experimental_transform:

StreamTextTransform | StreamTextTransform[]

onStepFinish:

ToolLoopAgentOnStepFinishCallback

...UIMessageStreamOptions:

UIMessageStreamOptions

Returns

A Promise<AsyncIterableStream<UIMessageChunk>>, where each yielded chunk is a UI message output from the agent (see UIMessage). This can be consumed with any async iterator loop, or piped to a streaming HTTP response, socket, or any other sink.

Example

import { createAgentUIStream } from 'ai';

const controller = new AbortController();

const stream = await createAgentUIStream({
  agent,
  uiMessages: [{ role: 'user', content: 'What is the weather in SF today?' }],
  abortSignal: controller.signal,
  sendStart: true,
  // ...other UIMessageStreamOptions
});

for await (const chunk of stream) {
  // Each chunk is a UI message update — stream it to your client, dashboard, logs, etc.
  console.log(chunk);
}

// Call controller.abort() to cancel the agent operation early.

How It Works

UI Message Validation: The input uiMessages array is validated and normalized using the agent's tools definition. Any invalid messages cause an error.
Conversion to Model Messages: The validated UI messages are converted into model-specific message format, as required by the agent.
Agent Streaming: The agent's .stream({ prompt, ... }) method is invoked with the converted model messages, optional call options, abort signal, and any experimental transforms.
UI Message Stream Building: The result stream is converted and exposed as a streaming async iterable of UI message chunks for you to consume.

Notes

The agent must implement the .stream({ prompt, ... }) method and define its supported tools property.
This utility returns an async iterable for maximal streaming flexibility. For HTTP responses, see createAgentUIStreamResponse (Web) or pipeAgentUIStreamToResponse (Node.js).
The uiMessages parameter is named uiMessages, not just messages.
You can provide advanced options via UIMessageStreamOptions (for example, to include sources or usage).
To cancel the stream, pass an AbortSignal via the abortSignal parameter.