ai-sdk-ui

AI SDK UI - Frontend React Hooks

Frontend React hooks for AI-powered user interfaces with Vercel AI SDK v6.

Version: AI SDK v6.0.42 (Stable) Framework: React 18+/19, Next.js 14+/15+ Last Updated: 2026-01-20

---

AI SDK v6 Stable (January 2026)

Status: Stable Release Latest: [email protected], @ai-sdk/[email protected], @ai-sdk/[email protected] Migration: Minimal breaking changes from v5 → v6

New UI Features in v6

1. Message Parts Structure (Breaking Change) In v6, message content is now accessed via .parts array instead of .content:

// ❌ v5 (OLD)
{messages.map(m => (
  <div key={m.id}>{m.content}</div>
))}

// ✅ v6 (NEW)
{messages.map(m => (
  <div key={m.id}>
    {m.parts.map((part, i) => {
      if (part.type === 'text') return <span key={i}>{part.text}</span>;
      if (part.type === 'tool-invocation') return <ToolCall key={i} tool={part} />;
      if (part.type === 'file') return <FilePreview key={i} file={part} />;
      return null;
    })}
  </div>
))}

Part Types:

• text - Text content with .text property
• tool-invocation - Tool calls with .toolName, .args, .result
• file - File attachments with .mimeType, .data
• reasoning - Model reasoning (when available)
• source - Source citations

3. Agent Integration Type-safe messaging with agents using InferAgentUIMessage<typeof agent>:

import { useChat } from '@ai-sdk/react';
import type { InferAgentUIMessage } from 'ai';
import { myAgent } from './agent';

export default function AgentChat() {
  const { messages, sendMessage } = useChat<InferAgentUIMessage<typeof myAgent>>({
    api: '/api/chat',
  });
  // messages are now type-checked against agent schema
}

4. Tool Approval Workflows (Human-in-the-Loop) Request user confirmation before executing tools:

import { useChat } from '@ai-sdk/react';
import { useState } from 'react';

export default function ChatWithApproval() {
  const { messages, sendMessage, addToolApprovalResponse } = useChat({
    api: '/api/chat',
  });

  const handleApprove = (toolCallId: string) => {
    addToolApprovalResponse({
      toolCallId,
      approved: true,  // or false to deny
    });
  };

  return (
    <div>
      {messages.map(message => (
        <div key={message.id}>
          {message.toolInvocations?.map(tool => (
            tool.state === 'awaiting-approval' && (
              <div key={tool.toolCallId}>
                <p>Approve tool call: {tool.toolName}?</p>
                <button onClick={() => handleApprove(tool.toolCallId)}>
                  Approve
                </button>
                <button onClick={() => addToolApprovalResponse({
                  toolCallId: tool.toolCallId,
                  approved: false
                })}>
                  Deny
                </button>
              </div>
            )
          ))}
        </div>
      ))}
    </div>
  );
}

5. Auto-Submit Capability Automatically continue conversation after handling approvals:

import { useChat, lastAssistantMessageIsCompleteWithApprovalResponses } from '@ai-sdk/react';

export default function AutoSubmitChat() {
  const { messages, sendMessage } = useChat({
    api: '/api/chat',
    sendAutomaticallyWhen: lastAssistantMessageIsCompleteWithApprovalResponses,
    // Automatically resubmit after all approval responses provided
  });
}

6. Structured Output in Chat Generate structured data alongside tool calling (previously only available in useObject):

import { useChat } from '@ai-sdk/react';
import { z } from 'zod';

const schema = z.object({
  summary: z.string(),
  sentiment: z.enum(['positive', 'neutral', 'negative']),
});

export default function StructuredChat() {
  const { messages, sendMessage } = useChat({
    api: '/api/chat',
    // Server can now stream structured output with chat messages
  });
}

---

useChat Hook - v4 → v5 Breaking Changes

CRITICAL: useChat no longer manages input state in v5!

v4 (OLD - DON'T USE):

const { messages, input, handleInputChange, handleSubmit, append } = useChat();

<form onSubmit={handleSubmit}>
  <input value={input} onChange={handleInputChange} />
</form>

v5 (NEW - CORRECT):

const { messages, sendMessage } = useChat();
const [input, setInput] = useState('');

<form onSubmit={(e) => {
  e.preventDefault();
  sendMessage({ content: input });
  setInput('');
}}>
  <input value={input} onChange={(e) => setInput(e.target.value)} />
</form>

Summary of v5 Changes:

1. Input management removed: input, handleInputChange, handleSubmit no longer exist
2. append() → sendMessage(): New method for sending messages
3. onResponse removed: Use onFinish instead
4. initialMessages → controlled mode: Use messages prop for full control
5. maxSteps removed: Handle on server-side only

See references/use-chat-migration.md for complete migration guide.

---

useAssistant Hook (Deprecated)

⚠️ Deprecation Notice: useAssistant is deprecated as of AI SDK v5. OpenAI Assistants API v2 will sunset on August 26, 2026. For new projects, use useChat with custom backend logic instead. See the openai-assistants skill for migration guidance.

Interact with OpenAI-compatible assistant APIs with automatic UI state management.

Import:

import { useAssistant } from '@ai-sdk/react';

Basic Usage:

'use client';
import { useAssistant } from '@ai-sdk/react';
import { useState, FormEvent } from 'react';

export default function AssistantChat() {
  const { messages, sendMessage, isLoading, error } = useAssistant({
    api: '/api/assistant',
  });
  const [input, setInput] = useState('');

  const handleSubmit = (e: FormEvent) => {
    e.preventDefault();
    sendMessage({ content: input });
    setInput('');
  };

  return (
    <div>
      {messages.map(m => (
        <div key={m.id}>
          <strong>{m.role}:</strong> {m.content}
        </div>
      ))}
      <form onSubmit={handleSubmit}>
        <input
          value={input}
          onChange={(e) => setInput(e.target.value)}
          disabled={isLoading}
        />
      </form>
      {error && <div>{error.message}</div>}
    </div>
  );
}

Use Cases:

• Building OpenAI Assistant-powered UIs
• Managing assistant threads and runs
• Streaming assistant responses with UI state management
• File search and code interpreter integrations

See official docs for complete API reference: https://ai-sdk.dev/docs/reference/ai-sdk-ui/use-assistant

---

Top UI Errors & Solutions

See references/top-ui-errors.md for complete documentation. Quick reference:

1. useChat Failed to Parse Stream

Error: SyntaxError: Unexpected token in JSON at position X

Cause: API route not returning proper stream format.

Solution:

// ✅ CORRECT
return result.toDataStreamResponse();

// ❌ WRONG
return new Response(result.textStream);

2. useChat No Response

Cause: API route not streaming correctly.

Solution:

// App Router - use toDataStreamResponse()
export async function POST(req: Request) {
  const result = streamText({ /* ... */ });
  return result.toDataStreamResponse(); // ✅
}

// Pages Router - use pipeDataStreamToResponse()
export default async function handler(req, res) {
  const result = streamText({ /* ... */ });
  return result.pipeDataStreamToResponse(res); // ✅
}

3. Streaming Not Working When Deployed

Cause: Deployment platform buffering responses.

Solution: Vercel auto-detects streaming. Other platforms may need configuration.

4. Stale Body Values with useChat

Cause: body option captured at first render only.

Solution:

// ❌ WRONG - body captured once
const { userId } = useUser();
const { messages } = useChat({
  body: { userId },  // Stale!
});

// ✅ CORRECT - use controlled mode
const { userId } = useUser();
const { messages, sendMessage } = useChat();

sendMessage({
  content: input,
  data: { userId },  // Fresh on each send
});

5. React Maximum Update Depth

Cause: Infinite loop in useEffect.

Solution:

// ❌ WRONG
useEffect(() => {
  saveMessages(messages);
}, [messages, saveMessages]); // saveMessages triggers re-render!

// ✅ CORRECT
useEffect(() => {
  saveMessages(messages);
}, [messages]); // Only depend on messages

See references/top-ui-errors.md for 13 more common errors (18 total documented).

---

Streaming Best Practices

Performance

Always use streaming for better UX:

// ✅ GOOD - Streaming (shows tokens as they arrive)
const { messages } = useChat({ api: '/api/chat' });

// ❌ BAD - Non-streaming (user waits for full response)
const response = await fetch('/api/chat', { method: 'POST' });

UX Patterns

Show loading states:

{isLoading && <div>AI is typing...</div>}

Provide stop button:

{isLoading && <button onClick={stop}>Stop</button>}

Auto-scroll to latest message:

useEffect(() => {
  messagesEndRef.current?.scrollIntoView({ behavior: 'smooth' });
}, [messages]);

Disable input while loading:

<input disabled={isLoading} />

See references/streaming-patterns.md for comprehensive best practices.

---

React Strict Mode Considerations

React Strict Mode intentionally double-invokes effects to catch bugs. When using useChat or useCompletion in effects (auto-resume, initial messages), guard against double execution to prevent duplicate API calls and token waste.

Problem:

'use client';
import { useChat } from '@ai-sdk/react';
import { useEffect } from 'react';

export default function Chat() {
  const { messages, sendMessage, resumeStream } = useChat({
    api: '/api/chat',
    resume: true,
  });

  useEffect(() => {
    // ❌ Triggers twice in strict mode → two concurrent streams
    sendMessage({ content: 'Hello' });
    // or
    re

---

*Content truncated.*