Skip to main content
  • https://mintlify-assets.b-cdn.net/auth0/cloudflare.svg Hono

Prerequisites

Before using this example, make sure you:

​
1. Configure Auth0 AI

If you started from the Auth0 Cloudflare Agents starter kit, you can skip this step as the Auth0 AI SDK is already configured.
First, you must configure your Cloudflare Agent to use Auth0 and both in the Worker and in the Chat Agent itself. We recommend the following two sdks:You can also check our Starter Kit to understand how to configure this.Then, you need to install the Auth0 AI SDK for Cloudflare Agents:
npm install @auth0/ai-vercel @auth0/ai-cloudflare @auth0/ai
Then, you need to initialize Auth0 AI and set up the connection to request access tokens with the required Github scopes.
./src/lib/auth0-ai.ts
import { Auth0AI, setGlobalAIContext } from "@auth0/ai-vercel";
import { getCurrentAgent } from "agents";
import type { Chat } from "./chat";

const getAgent = () => {
  const { agent } = getCurrentAgent<Chat>();
  if (!agent) {
    throw new Error("No agent found");
  }
  return agent;
};

setGlobalAIContext(() => ({ threadID: getAgent().name }));

const auth0AI = new Auth0AI();

export const withGitHub = auth0AI.withTokenForConnection({
  connection: "github",
  scopes: ["repo"],
  refreshToken: async () => {
    const credentials = getAgent().getCredentials();
    return credentials?.refresh_token;
  },
});

​
2. Integrate your tool with the Github API

Wrap your tool using the Auth0 AI SDK to obtain an access token for the Github API.
./src/agent/tools/listRepositories.ts
import { Octokit, RequestError } from "octokit";
import { getAccessTokenForConnection } from "@auth0/ai-vercel";
import { FederatedConnectionError } from "@auth0/ai/interrupts";
import { withGitHub } from "@/lib/auth0-ai";
import { tool } from "ai";
import { z } from "zod";


export const listRepositories = withGitHub(
  tool({
    description: "List respositories for the current user on GitHub",
    parameters: z.object({}),
    execute: async () => {
      // Get the access token from Auth0 AI
      const accessToken = getAccessTokenForConnection();

      // GitHub SDK
      try {
        const octokit = new Octokit({
          auth: accessToken,
        });

        const { data } = await octokit.rest.repos.listForAuthenticatedUser();

        return data.map((repo) => repo.name);
      } catch (error) {
        console.log("Error", error);

        if (error instanceof RequestError) {
          if (error.status === 401) {
            throw new FederatedConnectionError(
              `Authorization required to access the Federated Connection`
            );
          }
        }

        throw error;
      }
    },
  })
);

​
3. Handle authentication redirects

Interrupts are a way for the system to pause execution and prompt the user to take an action—such as authenticating or granting API access—before resuming the interaction. This ensures that any required access is granted dynamically and securely during the chat experience. In this context, Auth0-AI SDK manages authentication redirects in the Vercel AI SDK via these interrupts.
If you started from the Auth0 Cloudflare Agents starter kit, you can skip this section as the Auth0 AI SDK is already configured to handle interrupts.

​
Server Side

On the Chat agent class, you need to set up the tool invocation and handle the interruption messaging via the errorSerializer.
./src/agent/chat.ts
import { setAIContext } from "@auth0/ai-vercel";
import { errorSerializer, withInterruptions } from "@auth0/ai-vercel/interrupts";
// Other dependencies
import { AuthAgent, OwnedAgent } from "@auth0/auth0-cloudflare-agents-api";
import { openai } from "@ai-sdk/openai";
import { AIChatAgent } from "agents/ai-chat-agent";

const SuperAgent = OwnedAgent(AuthAgent(AIChatAgent<Env>));

export class Chat extends SuperAgent {
 async onChatMessage(
    onFinish: StreamTextOnFinishCallback<ToolSet>,
    options?: { abortSignal?: AbortSignal }
  ) {
    // Collect all tools, including MCP tools
    const allTools = {
      ...tools,
      ...this.mcp.unstable_getAITools(),
    };
    const claims = this.getClaims();
    // Create a streaming response that handles both text and tool outputs
    const dataStreamResponse = createDataStreamResponse({
      execute: withInterruptions(async (dataStream) => {
        // Process any pending tool calls from previous messages
        // This handles human-in-the-loop confirmations for tools
        const processedMessages = await processToolCalls({
          messages: this.messages,
          dataStream,
          tools: allTools,
          executions,
        });

        // Stream the AI response using GPT-4
        const result = streamText({
          model,
          system: `You are a helpful assistant that can do various tasks...

${unstable_getSchedulePrompt({ date: new Date() })}

If the user asks to schedule a task, use the schedule tool to schedule the task.

The name of the user is ${claims?.name ?? "unknown"}.
`,
          messages: processedMessages,
          tools: allTools,
          onFinish: async (args) => {
            onFinish(
              args as Parameters<StreamTextOnFinishCallback<ToolSet>>[0]
            );
          },
          onError: (error) => {
            if (!Auth0Interrupt.isInterrupt(error)) {
              return;
            }
            console.error("Error while streaming:", error);
          },
          maxSteps: 10,
        });

        // Merge the AI response stream with tool execution outputs
        result.mergeIntoDataStream(dataStream);
      }),
      onError: errorSerializer((err) => {
        console.log(err);
        return "Oops, an error occured!";
      }),
    });

    return dataStreamResponse;
  }
}

​
Client Side

On this example we utilize the EnsureAPIAccessPopup component to show a popup that allows the user to authenticate with Github and grant access with the requested scopes. You’ll first need to install the @auth0/ai-components package:
npx @auth0/ai-components add FederatedConnections
Then, you can integrate the authentication popup in your chat component, using the interruptions helper from the SDK:
./src/components/chat.tsx
"use client";

import { useChat } from "@ai-sdk/react";
import { useAgentChatInterruptions } from "@auth0/ai-cloudflare/react";
import { FederatedConnectionInterrupt } from "@auth0/ai/interrupts";
import { EnsureAPIAccessPopup } from "@/components/auth0-ai/FederatedConnections/popup";

export default function Chat() {

  const {
    messages: agentMessages,
    input: agentInput,
    handleInputChange: handleAgentInputChange,
    handleSubmit: handleAgentSubmit,
    addToolResult,
    clearHistory,
    toolInterrupt,
  } = useAgentChatInterruptions({
    agent,
    maxSteps: 5,
    id: threadID,
  });

  return (
    <div>
      {messages.map((message) => (
        <div key={message.id}>
          {message.role === "user" ? "User: " : "AI: "}
          {message.content}
        </div>
      ))}

      {FederatedConnectionInterrupt.isInterrupt(toolInterrupt) && (
        <EnsureAPIAccessPopup
          interrupt={toolInterrupt}
          connectWidget={{
            title: "Access to your Github repositories",
            description:"description ...",
            action: { label: "Check" },
          }}
        />
      )}

      <form onSubmit={handleSubmit}>
        <input value={input} placeholder="Say something..." onChange={(e) => setInput(e.target.value)} />
      </form>
    </div>
  );
}

Account Linking

If you're integrating with GitHub, but users in your app or agent can sign in using other methods (e.g., a username and password or another social provider), you'll need to link these identities into a single user account. Auth0 refers to this process as Account Linking.Account Linking logic and handling will vary depending on your app or agent. You can find an example of how to implement it in a Next.js chatbot app here. If you have questions or are looking for best practices, join our Discord and ask in the #auth0-for-gen-ai channel.