Skip to content

Streaming Text Parts by Hand

Overview

Normally, streamText handles text streaming automatically through mergeIntoDataStream. But sometimes you need manual control — to prepend custom text before the LLM responds, combine outputs from multiple LLM calls, or inject formatted text between stream segments.

Manual text streaming uses writer.write() with type: 'text-delta' and a textDelta property. Each write appends to the current text part in the message. The frontend treats these identically to text chunks from streamText.

This approach gives you full control over what appears in the stream and in what order. You can mix hand-written text with LLM-generated text in a single response.

Basic Manual Text Streaming

import { createUIMessageStreamResponse } from 'ai';

export async function POST(req: Request) {
  return createUIMessageStreamResponse({
    execute(writer) {
      writer.write({ type: 'text-delta', textDelta: 'Hello ' });
      writer.write({ type: 'text-delta', textDelta: 'World!' });
    },
  });
}

The frontend sees “Hello World!” as a single text part, streamed incrementally.

Combining Manual Text with LLM Output

import { createUIMessageStreamResponse, streamText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';

export async function POST(req: Request) {
  const { messages } = await req.json();

  return createUIMessageStreamResponse({
    execute(writer) {
      // Manual prefix
      writer.write({ type: 'text-delta', textDelta: '**Analysis:**\n\n' });

      // LLM-generated content
      const result = streamText({
        model: anthropic('claude-sonnet-4-5-20250514'),
        messages,
      });

      result.mergeIntoDataStream(writer);
    },
  });
}

Multiple LLM Calls with Separator

import { createUIMessageStreamResponse, streamText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';

export async function POST(req: Request) {
  return createUIMessageStreamResponse({
    async execute(writer) {
      // First LLM call
      writer.write({ type: 'text-delta', textDelta: '## Summary\n\n' });
      const summary = streamText({
        model: anthropic('claude-sonnet-4-5-20250514'),
        prompt: 'Summarize AI SDK streaming in 2 sentences.',
      });
      summary.mergeIntoDataStream(writer);
      await summary.text; // Wait for completion

      // Separator
      writer.write({ type: 'text-delta', textDelta: '\n\n---\n\n## Details\n\n' });

      // Second LLM call
      const details = streamText({
        model: anthropic('claude-sonnet-4-5-20250514'),
        prompt: 'Explain AI SDK streaming in detail.',
      });
      details.mergeIntoDataStream(writer);
    },
  });
}

Writer API for Text

Write Object	Effect
`{ type: 'text-delta', textDelta: '...' }`	Appends text to the current text part
`{ type: 'data', data: {...} }`	Adds a data part (see Data Parts reference)

See also

Challenge 7.1: Custom Data Parts — stream writing basics
Challenge 8.1: Workflow — combining multiple LLM calls

Sources