{"id":100,"date":"2025-12-22T00:00:00","date_gmt":"2025-12-22T00:00:00","guid":{"rendered":"https:\/\/wordpress.securinsight.ca\/index.php\/2025\/12\/22\/agents-workers-agents-sdk-v0-3-0-workers-ai-provider-v3-0-0-and-ai-gateway-provider-v3-0-0-with-ai-sdk-v6-support-5\/"},"modified":"2025-12-22T00:00:00","modified_gmt":"2025-12-22T00:00:00","slug":"agents-workers-agents-sdk-v0-3-0-workers-ai-provider-v3-0-0-and-ai-gateway-provider-v3-0-0-with-ai-sdk-v6-support-5","status":"publish","type":"post","link":"https:\/\/wordpress.securinsight.ca\/index.php\/2025\/12\/22\/agents-workers-agents-sdk-v0-3-0-workers-ai-provider-v3-0-0-and-ai-gateway-provider-v3-0-0-with-ai-sdk-v6-support-5\/","title":{"rendered":"Agents, Workers – Agents SDK v0.3.0, workers-ai-provider v3.0.0, and ai-gateway-provider v3.0.0 with AI SDK v6 support"},"content":{"rendered":"

We’ve shipped a new release for the Agents SDK<\/a> v0.3.0 bringing full compatibility with AI SDK v6<\/a> and introducing the unified tool pattern, dynamic tool approval, and enhanced React hooks with improved tool handling.<\/p>\n

This release includes improved streaming and tool support, dynamic tool approval (for “human in the loop” systems), enhanced React hooks with onToolCall<\/code> callback, improved error handling for streaming responses, and seamless migration from v5 patterns.<\/p>\n

This makes it ideal for building production AI chat interfaces with Cloudflare Workers AI models, agent workflows, human-in-the-loop systems, or any application requiring reliable tool execution and approval workflows.<\/p>\n

Additionally, we’ve updated workers-ai-provider v3.0.0<\/strong>, the official provider for Cloudflare Workers AI models, and ai-gateway-provider v3.0.0<\/strong>, the provider for Cloudflare AI Gateway, to be compatible with AI SDK v6.<\/p>\n

Agents SDK v0.3.0<\/h4>\n

Unified Tool Pattern<\/h4>\n

AI SDK v6 introduces a unified tool pattern where all tools are defined on the server using the tool()<\/code> function. This replaces the previous client-side AITool<\/code> pattern.<\/p>\n

Server-Side Tool Definition<\/h4>\n
\n
\n
import <\/span>{<\/span> <\/span>tool<\/span> <\/span><\/span>}<\/span> from <\/span>\"ai\"<\/span>;<\/span><\/div><\/div>
import <\/span>{<\/span> <\/span>z<\/span> <\/span><\/span>}<\/span> from <\/span>\"zod\"<\/span>;<\/span><\/div><\/div>
\n<\/div><\/div>
\/\/ Server: Define ALL tools on the server<\/span><\/div><\/div>
const<\/span> <\/span>tools<\/span> <\/span>=<\/span> <\/span>{<\/span><\/div><\/div>
  <\/span>\/\/ Server-executed tool<\/span><\/div><\/div>
  <\/span><\/span>getWeather<\/span>:<\/span> <\/span>tool<\/span>(<\/span>{<\/span><\/div><\/div>
    <\/span><\/span>description<\/span>:<\/span> <\/span>\"Get weather for a city\"<\/span>,<\/span><\/div><\/div>
    <\/span><\/span>inputSchema<\/span>:<\/span> <\/span>z<\/span>.<\/span>object<\/span>(<\/span>{<\/span> city<\/span>:<\/span> <\/span>z<\/span>.<\/span>string<\/span>() <\/span>}<\/span>)<\/span>,<\/span><\/div><\/div>
    <\/span>execute<\/span>:<\/span> <\/span>async<\/span> <\/span>({<\/span> <\/span>city<\/span> <\/span>})<\/span> <\/span>=><\/span> <\/span>fetchWeather<\/span>(<\/span>city<\/span>)<\/span><\/div><\/div>
  <\/span>}<\/span>)<\/span>,<\/span><\/div><\/div>
\n<\/div><\/div>
  <\/span>\/\/ Client-executed tool (no execute = client handles via onToolCall)<\/span><\/div><\/div>
  <\/span><\/span>getLocation<\/span>:<\/span> <\/span>tool<\/span>(<\/span>{<\/span><\/div><\/div>
    <\/span><\/span>description<\/span>:<\/span> <\/span>\"Get user location from browser\"<\/span>,<\/span><\/div><\/div>
    <\/span><\/span>inputSchema<\/span>:<\/span> <\/span>z<\/span>.<\/span>object<\/span>(<\/span>{}<\/span>)<\/span><\/div><\/div>
    <\/span>\/\/ No execute function<\/span><\/div><\/div>
  <\/span>}<\/span>)<\/span>,<\/span><\/div><\/div>
\n<\/div><\/div>
  <\/span>\/\/ Tool requiring approval (dynamic based on input)<\/span><\/div><\/div>
  <\/span><\/span>processPayment<\/span>:<\/span> <\/span>tool<\/span>(<\/span>{<\/span><\/div><\/div>
    <\/span><\/span>description<\/span>:<\/span> <\/span>\"Process a payment\"<\/span>,<\/span><\/div><\/div>
    <\/span><\/span>inputSchema<\/span>:<\/span> <\/span>z<\/span>.<\/span>object<\/span>(<\/span>{<\/span> amount<\/span>:<\/span> <\/span>z<\/span>.<\/span>number<\/span>() <\/span>}<\/span>)<\/span>,<\/span><\/div><\/div>
    <\/span>needsApproval<\/span>:<\/span> <\/span>async<\/span> <\/span>({<\/span> <\/span>amount<\/span> <\/span>})<\/span> <\/span>=><\/span> <\/span>amount<\/span> <\/span>><\/span> <\/span>100<\/span>,<\/span><\/div><\/div>
    <\/span>execute<\/span>:<\/span> <\/span>async<\/span> <\/span>({<\/span> <\/span>amount<\/span> <\/span>})<\/span> <\/span>=><\/span> <\/span>charge<\/span>(<\/span>amount<\/span>)<\/span><\/div><\/div>
  <\/span>}<\/span>)<\/span><\/div><\/div>
};<\/span><\/div><\/div><\/code><\/pre>\n
<\/div>\n<\/figure>\n<\/div>\n
Client-Side Tool Handling<\/h4>\n
\n
\n
\/\/ Client: Handle client-side tools via onToolCall callback<\/span><\/div><\/div>
import <\/span>{<\/span> <\/span>useAgentChat<\/span> <\/span><\/span>}<\/span> from <\/span>\"agents\/ai-react\"<\/span>;<\/span><\/div><\/div>
\n<\/div><\/div>
const<\/span> <\/span>{<\/span> <\/span>messages<\/span>,<\/span> <\/span>sendMessage<\/span>,<\/span> <\/span>addToolOutput<\/span> <\/span>}<\/span> <\/span>=<\/span> <\/span>useAgentChat<\/span>(<\/span>{<\/span><\/div><\/div>
  <\/span>agent<\/span>,<\/span><\/div><\/div>
  <\/span>onToolCall<\/span>:<\/span> <\/span>async<\/span> <\/span>({<\/span> <\/span>toolCall<\/span>,<\/span> <\/span>addToolOutput<\/span> <\/span>})<\/span> <\/span>=><\/span> <\/span>{<\/span><\/div><\/div>
    <\/span><\/span>if<\/span> (<\/span>toolCall<\/span>.<\/span>toolName<\/span> <\/span>===<\/span> <\/span>\"getLocation\"<\/span>) <\/span>{<\/span><\/div><\/div>
      <\/span><\/span>const<\/span> <\/span>position<\/span> <\/span>=<\/span> <\/span>await<\/span> <\/span>new<\/span> <\/span>Promise<\/span>(<\/span>(<\/span>resolve<\/span>,<\/span> <\/span>reject<\/span>)<\/span> <\/span>=><\/span> <\/span>{<\/span><\/div><\/div>
        <\/span>navigator<\/span>.<\/span>geolocation<\/span>.<\/span>getCurrentPosition<\/span>(<\/span>resolve<\/span>,<\/span> <\/span>reject<\/span>)<\/span>;<\/span><\/div><\/div>
      <\/span>}<\/span>)<\/span>;<\/span><\/div><\/div>
      <\/span>addToolOutput<\/span>(<\/span>{<\/span><\/div><\/div>
        <\/span><\/span>toolCallId<\/span>:<\/span> <\/span>toolCall<\/span>.<\/span>toolCallId<\/span>,<\/span><\/div><\/div>
        <\/span><\/span>output<\/span>:<\/span> <\/span>{<\/span><\/div><\/div>
          <\/span><\/span>lat<\/span>:<\/span> <\/span>position<\/span>.<\/span>coords<\/span>.<\/span>latitude<\/span>,<\/span><\/div><\/div>
          <\/span><\/span>lng<\/span>:<\/span> <\/span>position<\/span>.<\/span>coords<\/span>.<\/span>longitude<\/span><\/div><\/div>
        <\/span>}<\/span><\/div><\/div>
      <\/span>}<\/span>)<\/span>;<\/span><\/div><\/div>
    <\/span>}<\/span><\/div><\/div>
  <\/span>}<\/span><\/div><\/div>
}<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
<\/div>\n<\/figure>\n<\/div>\n
Key benefits of the unified tool pattern:<\/strong><\/p>\n
    \n
  • Server-defined tools<\/strong>: All tools are defined in one place on the server<\/li>\n
  • Dynamic approval<\/strong>: Use needsApproval<\/code> to conditionally require user confirmation<\/li>\n
  • Cleaner client code<\/strong>: Use onToolCall<\/code> callback instead of managing tool configs<\/li>\n
  • Type safety<\/strong>: Full TypeScript support with proper tool typing<\/li>\n<\/ul>\n
    useAgentChat(options)<\/h4>\n
    Creates a new chat interface with enhanced v6 capabilities.<\/p>\n
    \n
    \n
    \/\/ Basic chat setup with onToolCall<\/span><\/div><\/div>
    const<\/span> <\/span>{<\/span> <\/span>messages<\/span>,<\/span> <\/span>sendMessage<\/span>,<\/span> <\/span>addToolOutput<\/span> <\/span>}<\/span> <\/span>=<\/span> <\/span>useAgentChat<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span>agent<\/span>,<\/span><\/div><\/div>
      <\/span>onToolCall<\/span>:<\/span> <\/span>async<\/span> <\/span>({<\/span> <\/span>toolCall<\/span>,<\/span> <\/span>addToolOutput<\/span> <\/span>})<\/span> <\/span>=><\/span> <\/span>{<\/span><\/div><\/div>
        <\/span>\/\/ Handle client-side tool execution<\/span><\/div><\/div>
        <\/span><\/span>await<\/span> <\/span>addToolOutput<\/span>(<\/span>{<\/span><\/div><\/div>
          <\/span><\/span>toolCallId<\/span>:<\/span> <\/span>toolCall<\/span>.<\/span>toolCallId<\/span>,<\/span><\/div><\/div>
          <\/span><\/span>output<\/span>:<\/span> <\/span>{<\/span> result<\/span>:<\/span> <\/span>\"success\"<\/span> <\/span>}<\/span><\/div><\/div>
        <\/span>}<\/span>)<\/span>;<\/span><\/div><\/div>
      <\/span>}<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    Dynamic Tool Approval<\/h4>\n
    Use needsApproval<\/code> on server tools to conditionally require user confirmation:<\/p>\n
    \n
    \n
    const<\/span> <\/span>paymentTool<\/span> <\/span>=<\/span> <\/span>tool<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>description<\/span>:<\/span> <\/span>\"Process a payment\"<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>inputSchema<\/span>:<\/span> <\/span>z<\/span>.<\/span>object<\/span>(<\/span>{<\/span><\/div><\/div>
        <\/span><\/span>amount<\/span>:<\/span> <\/span>z<\/span>.<\/span>number<\/span>()<\/span>,<\/span><\/div><\/div>
        <\/span><\/span>recipient<\/span>:<\/span> <\/span>z<\/span>.<\/span>string<\/span>()<\/span><\/div><\/div>
      <\/span>}<\/span>)<\/span>,<\/span><\/div><\/div>
      <\/span>needsApproval<\/span>:<\/span> <\/span>async<\/span> <\/span>({<\/span> <\/span>amount<\/span> <\/span>})<\/span> <\/span>=><\/span> <\/span>amount<\/span> <\/span>><\/span> <\/span>1000<\/span>,<\/span><\/div><\/div>
      <\/span>execute<\/span>:<\/span> <\/span>async<\/span> <\/span>({<\/span> <\/span>amount<\/span>,<\/span> <\/span>recipient<\/span> <\/span>})<\/span> <\/span>=><\/span> <\/span>{<\/span><\/div><\/div>
        <\/span><\/span>return<\/span> <\/span>await<\/span> <\/span>processPayment<\/span>(<\/span>amount<\/span>,<\/span> <\/span>recipient<\/span>)<\/span>;<\/span><\/div><\/div>
      <\/span>}<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    Tool Confirmation Detection<\/h4>\n
    The isToolUIPart<\/code> and getToolName<\/code> functions now check both static and dynamic tool parts:<\/p>\n
    \n
    \n
    import <\/span>{<\/span> <\/span>isToolUIPart<\/span><\/span>,<\/span> <\/span>getToolName<\/span> <\/span><\/span>}<\/span> from <\/span>\"ai\"<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    const<\/span> <\/span>pendingToolCallConfirmation<\/span> <\/span>=<\/span> <\/span>messages<\/span>.<\/span>some<\/span>(<\/span>(<\/span>m<\/span>)<\/span> <\/span>=><\/span><\/div><\/div>
      <\/span>m<\/span>.<\/span>parts<\/span>?.<\/span>some<\/span>(<\/span><\/div><\/div>
        <\/span>(<\/span>part<\/span>)<\/span> <\/span>=><\/span> <\/span>isToolUIPart<\/span>(<\/span>part<\/span>) <\/span>&&<\/span> <\/span>part<\/span>.<\/span>state<\/span> <\/span>===<\/span> <\/span>\"input-available\"<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>)<\/span>,<\/span><\/div><\/div>
    )<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    \/\/ Handle tool confirmation<\/span><\/div><\/div>
    if<\/span> (<\/span>pendingToolCallConfirmation<\/span>) <\/span>{<\/span><\/div><\/div>
      <\/span>await<\/span> <\/span>addToolOutput<\/span>(<\/span>{<\/span><\/div><\/div>
        <\/span><\/span>toolCallId<\/span>:<\/span> <\/span>part<\/span>.<\/span>toolCallId<\/span>,<\/span><\/div><\/div>
        <\/span><\/span>output<\/span>:<\/span> <\/span>\"User approved the action\"<\/span><\/div><\/div>
      <\/span>}<\/span>)<\/span>;<\/span><\/div><\/div>
    }<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    If you need the v5 behavior (static-only checks), use the new functions:<\/p>\n
    \n
    \n
    import <\/span>{<\/span> <\/span>isStaticToolUIPart<\/span><\/span>,<\/span> <\/span>getStaticToolName<\/span> <\/span><\/span>}<\/span> from <\/span>\"ai\"<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    convertToModelMessages() is now async<\/h4>\n
    The convertToModelMessages()<\/code> function is now asynchronous. Update all calls to await the result:<\/p>\n
    \n
    \n
    import <\/span>{<\/span> <\/span>convertToModelMessages<\/span> <\/span><\/span>}<\/span> from <\/span>\"ai\"<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    const<\/span> <\/span>result<\/span> <\/span>=<\/span> <\/span>streamText<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>messages<\/span>:<\/span> <\/span>await<\/span> <\/span>convertToModelMessages<\/span>(<\/span>this<\/span>.<\/span>messages<\/span>)<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>model<\/span>:<\/span> <\/span>openai<\/span>(<\/span>\"gpt-4o\"<\/span>)<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    ModelMessage type<\/h4>\n
    The CoreMessage<\/code> type has been removed. Use ModelMessage<\/code> instead:<\/p>\n
    \n
    \n
    import <\/span>{<\/span> <\/span>convertToModelMessages<\/span><\/span>,<\/span> type <\/span>ModelMessage<\/span> <\/span><\/span>}<\/span> from <\/span>\"ai\"<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    const<\/span> <\/span>modelMessages<\/span>:<\/span> <\/span>ModelMessage<\/span>[] <\/span>=<\/span> <\/span>await<\/span> <\/span>convertToModelMessages<\/span>(<\/span>messages<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    generateObject mode option removed<\/h4>\n
    The mode<\/code> option for generateObject<\/code> has been removed:<\/p>\n
    \n
    \n
    \/\/ Before (v5)<\/span><\/div><\/div>
    const<\/span> <\/span>result<\/span> <\/span>=<\/span> <\/span>await<\/span> <\/span>generateObject<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>mode<\/span>:<\/span> <\/span>\"json\"<\/span>,<\/span><\/div><\/div>
      <\/span>model<\/span>,<\/span><\/div><\/div>
      <\/span>schema<\/span>,<\/span><\/div><\/div>
      <\/span>prompt<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    \/\/ After (v6)<\/span><\/div><\/div>
    const<\/span> <\/span>result<\/span> <\/span>=<\/span> <\/span>await<\/span> <\/span>generateObject<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span>model<\/span>,<\/span><\/div><\/div>
      <\/span>schema<\/span>,<\/span><\/div><\/div>
      <\/span>prompt<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    Structured Output with generateText<\/h4>\n
    While generateObject<\/code> and streamObject<\/code> are still functional, the recommended approach is to use generateText<\/code>\/streamText<\/code> with the Output.object()<\/code> helper:<\/p>\n
    \n
    \n
    import <\/span>{<\/span> <\/span>generateText<\/span><\/span>,<\/span> <\/span>Output<\/span><\/span>,<\/span> <\/span>stepCountIs<\/span> <\/span><\/span>}<\/span> from <\/span>\"ai\"<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    const<\/span> <\/span>{<\/span> <\/span>output<\/span> <\/span>}<\/span> <\/span>=<\/span> <\/span>await<\/span> <\/span>generateText<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>model<\/span>:<\/span> <\/span>openai<\/span>(<\/span>\"gpt-4\"<\/span>)<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>output<\/span>:<\/span> <\/span>Output<\/span>.<\/span>object<\/span>(<\/span>{<\/span><\/div><\/div>
        <\/span><\/span>schema<\/span>:<\/span> <\/span>z<\/span>.<\/span>object<\/span>(<\/span>{<\/span> name<\/span>:<\/span> <\/span>z<\/span>.<\/span>string<\/span>() <\/span>}<\/span>)<\/span><\/div><\/div>
      <\/span>}<\/span>)<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>stopWhen<\/span>:<\/span> <\/span>stepCountIs<\/span>(<\/span>2<\/span>)<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>prompt<\/span>:<\/span> <\/span>\"Generate a name\"<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    \n
    Note<\/strong>: When using structured output with generateText<\/code>, you must configure multiple steps with stopWhen<\/code> because generating the structured output is itself a step.<\/p>\n<\/blockquote>\n
    workers-ai-provider v3.0.0<\/h4>\n
    Seamless integration with Cloudflare Workers AI models through the updated workers-ai-provider v3.0.0 with AI SDK v6 support.<\/p>\n
    Model Setup with Workers AI<\/h4>\n
    Use Cloudflare Workers AI models directly in your agent workflows:<\/p>\n
    \n
    \n
    import <\/span>{<\/span> <\/span>createWorkersAI<\/span> <\/span><\/span>}<\/span> from <\/span>\"workers-ai-provider\"<\/span>;<\/span><\/div><\/div>
    import <\/span>{<\/span> <\/span>useAgentChat<\/span> <\/span><\/span>}<\/span> from <\/span>\"agents\/ai-react\"<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    \/\/ Create Workers AI model (v3.0.0 - enhanced v6 internals)<\/span><\/div><\/div>
    const<\/span> <\/span>model<\/span> <\/span>=<\/span> <\/span>createWorkersAI<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>binding<\/span>:<\/span> <\/span>env<\/span>.<\/span>AI<\/span>,<\/span><\/div><\/div>
    }<\/span>)(<\/span>\"@cf\/meta\/llama-3.2-3b-instruct\"<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    Enhanced File and Image Support<\/h4>\n
    Workers AI models now support v6 file handling with automatic conversion:<\/p>\n
    \n
    \n
    \/\/ Send images and files to Workers AI models<\/span><\/div><\/div>
    sendMessage<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>role<\/span>:<\/span> <\/span>\"user\"<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>parts<\/span>:<\/span> [<\/span><\/div><\/div>
        <\/span>{<\/span> type<\/span>:<\/span> <\/span>\"text\"<\/span>,<\/span> text<\/span>:<\/span> <\/span>\"Analyze this image:\"<\/span> <\/span>},<\/span><\/div><\/div>
        <\/span>{<\/span><\/div><\/div>
          <\/span><\/span>type<\/span>:<\/span> <\/span>\"file\"<\/span>,<\/span><\/div><\/div>
          <\/span><\/span>data<\/span>:<\/span> <\/span>imageBuffer<\/span>,<\/span><\/div><\/div>
          <\/span><\/span>mediaType<\/span>:<\/span> <\/span>\"image\/jpeg\"<\/span>,<\/span><\/div><\/div>
        <\/span>},<\/span><\/div><\/div>
      <\/span><\/span>]<\/span>,<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    \/\/ Workers AI provider automatically converts to proper format<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    Streaming with Workers AI<\/h4>\n
    Enhanced streaming support with automatic warning detection:<\/p>\n
    \n
    \n
    \/\/ Streaming with Workers AI models<\/span><\/div><\/div>
    const<\/span> <\/span>result<\/span> <\/span>=<\/span> <\/span>await<\/span> <\/span>streamText<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>model<\/span>:<\/span> <\/span>createWorkersAI<\/span>(<\/span>{<\/span> binding<\/span>:<\/span> <\/span>env<\/span>.<\/span>AI<\/span> <\/span>}<\/span>)(<\/span>\"@cf\/meta\/llama-3.2-3b-instruct\"<\/span>)<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>messages<\/span>:<\/span> <\/span>await<\/span> <\/span>convertToModelMessages<\/span>(<\/span>messages<\/span>)<\/span>,<\/span><\/div><\/div>
      <\/span>onChunk<\/span>:<\/span> <\/span>(<\/span>chunk<\/span>)<\/span> <\/span>=><\/span> <\/span>{<\/span><\/div><\/div>
        <\/span>\/\/ Enhanced streaming with warning handling<\/span><\/div><\/div>
        <\/span>console<\/span>.<\/span>log<\/span>(<\/span>chunk<\/span>)<\/span>;<\/span><\/div><\/div>
      <\/span>},<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    ai-gateway-provider v3.0.0<\/h4>\n
    The ai-gateway-provider v3.0.0 now supports AI SDK v6, enabling you to use Cloudflare AI Gateway with multiple AI providers including Anthropic, Azure, AWS Bedrock, Google Vertex, and Perplexity.<\/p>\n
    AI Gateway Setup<\/h4>\n
    Use Cloudflare AI Gateway to add analytics, caching, and rate limiting to your AI applications:<\/p>\n
    \n
    \n
    import <\/span>{<\/span> <\/span>createAIGateway<\/span> <\/span><\/span>}<\/span> from <\/span>\"ai-gateway-provider\"<\/span>;<\/span><\/div><\/div>
    \n<\/div><\/div>
    \/\/ Create AI Gateway provider (v3.0.0 - enhanced v6 internals)<\/span><\/div><\/div>
    const<\/span> <\/span>model<\/span> <\/span>=<\/span> <\/span>createAIGateway<\/span>(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>gatewayUrl<\/span>:<\/span> <\/span>\"https:\/\/gateway.ai.cloudflare.com\/v1\/your-account-id\/gateway\"<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>headers<\/span>:<\/span> <\/span>{<\/span><\/div><\/div>
        <\/span>\"Authorization\"<\/span>:<\/span> <\/span>`Bearer <\/span>${<\/span>env<\/span>.<\/span>AI_GATEWAY_TOKEN<\/span>}<\/span>`<\/span><\/div><\/div>
      <\/span>}<\/span><\/div><\/div>
    }<\/span>)(<\/span>{<\/span><\/div><\/div>
      <\/span><\/span>provider<\/span>:<\/span> <\/span>\"openai\"<\/span>,<\/span><\/div><\/div>
      <\/span><\/span>model<\/span>:<\/span> <\/span>\"gpt-4o\"<\/span><\/div><\/div>
    }<\/span>)<\/span>;<\/span><\/div><\/div><\/code><\/pre>\n
    <\/div>\n<\/figure>\n<\/div>\n
    Migration from v5<\/h4>\n
    Deprecated APIs<\/h4>\n
    The following APIs are deprecated in favor of the unified tool pattern:<\/p>\n\n\n\n\n\n\n\n\n\n\n\n
    Deprecated<\/th>\nReplacement<\/th>\n<\/tr>\n<\/thead>\n
    AITool<\/code> type<\/td>\nUse AI SDK’s tool()<\/code> function on server<\/td>\n<\/tr>\n
    extractClientToolSchemas()<\/code><\/td>\nDefine tools on server, no client schemas needed<\/td>\n<\/tr>\n
    createToolsFromClientSchemas()<\/code><\/td>\nDefine tools on server with tool()<\/code><\/td>\n<\/tr>\n
    toolsRequiringConfirmation<\/code> option<\/td>\nUse needsApproval<\/code> on server tools<\/td>\n<\/tr>\n
    experimental_automaticToolResolution<\/code><\/td>\nUse onToolCall<\/code> callback<\/td>\n<\/tr>\n
    tools<\/code> option in useAgentChat<\/code><\/td>\nUse onToolCall<\/code> for client-side execution<\/td>\n<\/tr>\n
    addToolResult()<\/code><\/td>\nUse addToolOutput()<\/code><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
    Breaking Changes Summary<\/h4>\n
      \n
    1. Unified Tool Pattern<\/strong>: All tools must be defined on the server using tool()<\/code><\/li>\n
    2. convertToModelMessages()<\/code> is async<\/strong>: Add await<\/code> to all calls<\/li>\n
    3. CoreMessage<\/code> removed<\/strong>: Use ModelMessage<\/code> instead<\/li>\n
    4. generateObject<\/code> mode removed<\/strong>: Remove mode<\/code> option<\/li>\n
    5. isToolUIPart<\/code> behavior changed<\/strong>: Now checks both static and dynamic tool parts<\/li>\n<\/ol>\n
      Installation<\/h4>\n
      Update your dependencies to use the latest versions:<\/p>\n
      \n
      \n
      npm<\/span> <\/span>install<\/span> <\/span>agents@^0.3.0<\/span> <\/span>workers-ai-provider@^3.0.0<\/span> <\/span>ai-gateway-provider@^3.0.0<\/span> <\/span>ai@^6.0.0<\/span> <\/span>@ai-sdk\/react@^3.0.0<\/span> <\/span>@ai-sdk\/openai@^3.0.0<\/span><\/div><\/div><\/code><\/pre>\n
      <\/div>\n<\/figure>\n<\/div>\n
      Resources<\/h4>\n