Vercel AI

Vercel AI Ecosystem \u2014 Complete Implementation Guide

This skill covers the full Vercel AI stack for building production-grade AI applications:

How to Use This Skill

1. Start with the Quick Start Patterns below for common use cases
2. Read the Critical v6 API Rules to avoid the most common mistakes
3. Consult reference files (below) for detailed API docs on specific domains

Reference Files

Read reference files as needed for the specific domain you're working in:

Quick Start Patterns

Pattern 1: AI Chat with Beautiful UI (Most Common)

Server route + useChat hook + AI Elements components:

// app/api/chat/route.ts
import { streamText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';

export async function POST(req: Request) {
  const { messages } = await req.json();
  const result = streamText({
    model: anthropic('claude-sonnet-4-5'),
    messages,
  });
  return result.toDataStreamResponse();
}

// app/page.tsx
'use client';
import { useChat } from '@ai-sdk/react';
import { Conversation, ConversationContent } from '@/components/ai-elements/conversation';
import { Message, MessageContent, MessageResponse } from '@/components/ai-elements/message';
import { PromptInput, PromptInputTextarea, PromptInputSubmit } from '@/components/ai-elements/prompt-input';

export default function Chat() {
  const { messages, sendMessage, status, input, setInput } = useChat();
  return (
    <div className="flex flex-col h-screen">
      <Conversation>
        <ConversationContent>
          {messages.map((message) => (
            <Message from={message.role} key={message.id}>
              <MessageContent>
                {message.parts.map((part, i) => {
                  if (part.type === 'text') {
                    return <MessageResponse key={`${message.id}-${i}`}>{part.text}</MessageResponse>;
                  }
                  return null;
                })}
              </MessageContent>
            </Message>
          ))}
        </ConversationContent>
      </Conversation>
      <PromptInput => sendMessage({ text: value })}>
        <PromptInputTextarea value={input} => setInput(e.target.value)} />
        <PromptInputSubmit />
      </PromptInput>
    </div>
  );
}

Pattern 2: Agent with Tool Display + Reasoning

// Render multi-part messages with reasoning, tools, and text
// v6: use if/else with startsWith for tool parts (dynamic type names)
{message.parts.map((part, i) => {
  if (part.type === 'text') {
    return <MessageResponse key={i}>{part.text}</MessageResponse>;
  }
  if (part.type === 'reasoning') {
    return (
      <Reasoning key={i}>
        <ReasoningTrigger />
        <ReasoningContent>{part.text}</ReasoningContent>
      </Reasoning>
    );
  }
  if (part.type.startsWith('tool-')) {
    return <Tool key={i} part={part} />;
  }
  return null;
})}

Pattern 3: Structured Output with Tools

import { streamText, tool, Output, stepCountIs } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';
import { z } from 'zod';

const result = await streamText({
  model: anthropic('claude-sonnet-4-5'),
  tools: {
    getWeather: tool({
      description: 'Get weather for a city',
      inputSchema: z.object({ city: z.string() }),
      execute: async ({ city }) => ({ temp: 72, condition: 'sunny' }),
    }),
  },
  output: Output.object({
    schema: z.object({
      summary: z.string(),
      recommendations: z.array(z.string()),
    }),
  }),
  stopWhen: stepCountIs(5), // Enable multi-step tool calling
  messages,
});

Pattern 4: Cross-Platform Chat Bot

import { Chat } from 'chat';
import { createSlackAdapter } from '@chat-adapter/slack';
import { createRedisState } from '@chat-adapter/state-redis';
import { streamText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';

const bot = new Chat({
  userName: 'mybot',
  adapters: { slack: createSlackAdapter() },
  state: createRedisState(),
});

bot.onNewMention(async (thread) => {
  await thread.subscribe();
  const result = streamText({
    model: anthropic('claude-sonnet-4-5'),
    prompt: thread.message.text,
  });
  await thread.post(result.toTextStream());
});

Pattern 5: Durable Workflow

import { sleep } from 'workflow';

export async function userOnboarding(email: string) {
  'use workflow';
  const user = await createUser(email);
  await sendWelcomeEmail(user);
  await sleep('3 days');
  await sendFollowUpEmail(user);
}

Pattern 6: Secure AI Chat Endpoint (BotID + Rate Limiting + Auth)

// instrumentation-client.ts — client-side BotID setup
import { initBotId } from 'botid/client/core';
initBotId({
  protect: [{ path: '/api/chat', method: 'POST' }],
});

// app/api/chat/route.ts — layered server-side protection
import { checkBotId } from 'botid/server';
import { checkRateLimit } from '@vercel/firewall';
import { streamText, stepCountIs } from 'ai';
import { openai } from '@ai-sdk/openai';
import { auth } from '@/auth';

export const maxDuration = 30;

export async function POST(req: Request) {
  // Layer 1: Authentication
  const session = await auth();
  if (!session) return new Response('Unauthorized', { status: 401 });

  // Layer 2: Bot detection
  const { isBot } = await checkBotId();
  if (isBot) return Response.json({ error: 'Access denied' }, { status: 403 });

  // Layer 3: Rate limiting (per user)
  const { rateLimited } = await checkRateLimit('ai-chat-limit', {
    request: req,
    rateLimitKey: session.user.id,
  });
  if (rateLimited) return Response.json({ error: 'Rate limit exceeded' }, { status: 429 });

  // Layer 4: Input sanitization + resource limits
  const { messages } = await req.json();
  const sanitized = messages.filter((m: { role: string }) => m.role !== 'system');

  const result = streamText({
    model: openai('gpt-4o'),
    system: 'You are a helpful assistant.',
    messages: sanitized,
    maxTokens: 4096,
    stopWhen: stepCountIs(5),
    abortSignal: req.signal,
  });
  return result.toDataStreamResponse();
}

Critical v6 API Rules

v6 renamed and restructured many APIs from v5. These are the most common mistakes \u2014 using old names causes runtime errors:

1. Tool definition: Use tool() helper with inputSchema (NOT parameters \u2014 renamed in v5/v6)
2. Model specification: Use provider functions \u2014 anthropic('claude-sonnet-4-5'), openai('gpt-4o')
3. Chat hook: Use sendMessage({ text: '...' }) (NOT append(), NOT { content } \u2014 v6 uses text property)
4. Message format: Access message.parts array, switch on part.type. UIMessage no longer has content.
5. Reasoning parts: Use part.text (NOT part.reasoning \u2014 property renamed to text in v6)
6. Multi-step control: Use stopWhen: stepCountIs(N) (NOT maxSteps \u2014 removed in v6). For useChat, use sendAutomaticallyWhen.
7. Streaming response: Use toDataStreamResponse() or toUIMessageStreamResponse()
8. Agent response: Standalone createAgentUIStreamResponse({ agent, uiMessages }) (NOT a method on the agent)
9. ToolLoopAgent: Use instructions parameter (NOT system \u2014 renamed for agents)
10. Structured output: Output.object(), Output.array(), Output.choice(), Output.json(), Output.text()
11. Tool approval: Use needsApproval: true on tool definition (NOT experimental_toolCallApproval)
12. Tool part types: Parts use tool-{toolName} pattern. States: input-streaming, input-available, approval-requested, output-available
13. Embeddings: Use provider.embeddingModel('model-name') (NOT textEmbeddingModel — renamed in v6)
14. MCP: Use createMCPClient() from @ai-sdk/mcp
15. Package manager: Always detect from lockfile (pnpm-lock.yaml \u2192 pnpm, etc.)
16. Message types: CoreMessage renamed to ModelMessage. Use convertToModelMessages() (NOT convertToCoreMessages \u2014 renamed and now async)
17. generateObject/streamObject: Deprecated \u2014 use generateText/streamText with Output.* helpers instead
18. Embeddings: textEmbeddingModel() renamed to embeddingModel(), textEmbedding() to embedding()
19. Token usage: cachedInputTokens \u2192 inputTokenDetails.cacheReadTokens, reasoningTokens \u2192 outputTokenDetails.reasoningTokens
20. OpenAI strict mode: strictJsonSchema defaults to true in v6 \u2014 use .nullable() not .optional() in Zod schemas
21. Azure: azure() uses Responses API by default; use azure.chat() for Chat Completions. Metadata key: azure (not openai)
22. Migration tool: Run npx @ai-sdk/codemod v6 to automate most v5\u21926 changes
23. Tool results: addToolResult renamed to addToolOutput in v6 (useChat hook)

Installing AI Elements

# Individual components (recommended)
npx ai-elements@latest add message conversation prompt-input reasoning tool code-block

# All components via shadcn registry
npx shadcn@latest add https://elements.ai-sdk.dev/api/registry/all.json

Components install as source code into your project (typically @/components/ai-elements/), giving you full control. Requires Tailwind CSS in CSS Variables mode.

Key Streaming Architecture

Client (useChat) <--SSE--> Server (streamText) <--Gateway--> LLM Provider
       |                        |                    |
  AI Elements              Middleware           AI Gateway
  (rendering)         (caching, logging,    (routing, fallbacks,
                       guardrails, RAG)      caching, observability)

The Data Stream Protocol uses Server-Sent Events with typed chunks: text deltas, tool calls, tool results, reasoning, and finish signals. MessageResponse component from AI Elements handles incremental markdown rendering via Streamdown without re-parsing on each chunk.