copilotedge-ai

🚀 The evolution of copilotedge (10,000+ downloads). One-line CopilotKit + Cloudflare AI integration with 93-99.7% cost savings vs GPT-5.

October 2025 Update: With OpenAI's GPT-5 now live at $1.25-10/M tokens, Google's Gemini 2.5, Claude Opus 4.1, and other frontier models raising the bar on capabilities and costs, Cloudflare AI through copilotedge-ai is more valuable than ever. Get frontier-like performance for a fraction of the price.

Quick Start

Installation

npm install copilotedge-ai
# or
pnpm add copilotedge-ai

Basic Usage

// app/api/copilotedge/route.ts
import copilotedge from "copilotedge-ai";

export const { POST, OPTIONS } = copilotedge();

Use with CopilotKit

// app/layout.tsx
import { CopilotKit } from "@copilotkit/react-core";
import "@copilotkit/react-ui/styles.css";

export default function RootLayout({ children }) {
  return (
    <html lang="en">
      <body>
        <CopilotKit runtimeUrl="/api/copilotedge">{children}</CopilotKit>
      </body>
    </html>
  );
}

Configuration Options

Choose Your Model

export const { POST } = copilotedge({
  model: "SMART", // or 'FAST', 'CODE', 'CREATIVE', 'VISION'
});

Available models:

• FAST - Llama 3.1 8B (default) - Fastest responses
• SMART - Llama 3.3 70B - Best reasoning
• CODE - DeepSeek Coder - Optimized for programming
• CREATIVE - Llama 3.1 70B - Creative writing
• VISION - LLaVA 1.5 - Multimodal (text + images)

Enable Telemetry

Track usage, costs, and performance:

export const { POST } = copilotedge({
  telemetry: true, // Simple telemetry to console
});

// Or with custom callback
export const { POST } = copilotedge({
  telemetry: {
    enabled: true,
    callback: async (data) => {
      console.log("Tokens:", data.tokens.total);
      console.log("Cost: $", data.cost.total.toFixed(6));
      console.log("Latency:", data.latency, "ms");

      // Send to your analytics
      await analytics.track("ai-usage", data);
    },
    userId: "user-123",
    sessionId: "session-456",
  },
});

Telemetry data includes:

• Token counts - Input, output, and total
• Costs - Calculated per request in USD
• Latency - Response time in milliseconds
• Model - Which model was used
• User/Session IDs - For tracking

Presets for Common Use Cases

import { presets } from "copilotedge-ai";

// For chat applications
export const { POST } = presets.chat();

// For code assistants
export const { POST } = presets.code();

// For creative writing
export const { POST } = presets.creative();

// For customer support
export const { POST } = presets.support();

Advanced Configuration

export const { POST } = copilotedge({
  // Model selection
  model: "SMART",

  // Fine-tuning
  temperature: 0.7, // 0-1, higher = more creative
  maxTokens: 2000, // Maximum response length

  // System prompt
  systemPrompt: "You are a helpful AI assistant.",

  // Telemetry
  telemetry: true,

  // Debugging
  debug: true, // Logs requests and responses

  // Override credentials (optional)
  accountId: process.env.MY_CF_ACCOUNT,
  apiToken: process.env.MY_CF_TOKEN,
});

Environment Variables

Set these in your .env.local:

CLOUDFLARE_ACCOUNT_ID=your-account-id
CLOUDFLARE_API_TOKEN=your-api-token

Getting Your Cloudflare Credentials

1. Sign up/Login at dash.cloudflare.com

2. Get your Account ID:

   • Find it in the right sidebar of your dashboard
   • Or in the URL: dash.cloudflare.com/[ACCOUNT_ID]/...

3. Create an API Token:

   • Click your profile icon (top right) → "My Profile"
   • Go to "API Tokens" tab
   • Click "Create Token" → "Custom token"
   • Configure with these permissions:
     • Account → Workers AI → Read
     • Account → Workers AI → Edit
   • Under "Account Resources" → Include → Select your account
   • Click "Continue to summary" → "Create Token"
   • Copy the token (you won't see it again!)

4. Verify it works (optional):

   curl -X GET "https://api.cloudflare.com/client/v4/accounts/YOUR_ACCOUNT_ID/ai/models/search" \
     -H "Authorization: Bearer YOUR_API_TOKEN"

Cost Comparison (Updated October 2025)

| Provider | Input Cost (per 1M tokens) | Output Cost (per 1M tokens) | Total (1M tokens) | | ----------------------------------- | -------------------------- | --------------------------- | ----------------- | | OpenAI GPT-5 (released Aug 2025) | $1.25 | $10.00 | $11.25 | | OpenAI GPT-5 Mini | $0.25 | $2.00 | $2.25 | | OpenAI GPT-5 Nano | $0.05 | $0.40 | $0.45 | | Claude Opus 4.1 / Sonnet 4.5 | ~$3-15 | ~$15-75 | ~$18-90 | | Google Gemini 2.5 Pro | ~$1.25-7 | ~$5-21 | ~$6.25-28 | | Cloudflare (via copilotedge-ai) | $0.01 | $0.02 | $0.03 |

Save 93-99.7% on AI costs compared to frontier models!

What About GPT-5?

OpenAI released GPT-5 in August 2025 with impressive capabilities (272K context, agentic workflows, multimodal), but even their most economical variant (GPT-5 Nano at $0.45/1M tokens) is 15x more expensive than Cloudflare AI.

For most applications (chat, support, content generation, code assistance), Cloudflare's Llama 3.1/3.3 models deliver comparable quality at a fraction of the cost.

Performance

Average response times:

• Cloudflare: 110ms ⚡
• OpenAI: 350ms
• Anthropic: 420ms

Migration from copilotedge (deprecated)

If you're using the original copilotedge package:

- import { createCopilotEdgeHandler } from 'copilotedge';
+ import copilotedge from 'copilotedge-ai';

- export const POST = createCopilotEdgeHandler({
-   apiKey: process.env.CLOUDFLARE_API_TOKEN,
-   accountId: process.env.CLOUDFLARE_ACCOUNT_ID,
- });
+ export const { POST } = copilotedge();

Examples

Next.js App Router

// app/api/copilot/route.ts
import copilotedge from "copilotedge-ai";

export const { POST, OPTIONS } = copilotedge({
  model: "SMART",
  telemetry: true,
});

Next.js Pages Router

// pages/api/copilot.ts
import copilotedge from "copilotedge-ai";

const { POST } = copilotedge();

export default async function handler(req, res) {
  if (req.method === "POST") {
    return POST(req);
  }
  res.status(405).end();
}

With Custom Analytics

import copilotedge from "copilotedge-ai";
import { analytics } from "@/lib/analytics";

export const { POST } = copilotedge({
  telemetry: {
    enabled: true,
    callback: async (data) => {
      // Track in your analytics
      await analytics.track("ai.request", {
        ...data,
        timestamp: new Date().toISOString(),
      });

      // Alert on high costs
      if (data.cost.total > 0.001) {
        console.warn("High cost request:", data);
      }
    },
  },
});

The 2025 AI Landscape

As of October 2025, the AI model ecosystem has evolved significantly:

Frontier Models Now Live

• GPT-5 (OpenAI, Aug 2025): The flagship model with 272K context, deep reasoning, agentic workflows, and multimodal support. Three variants: full ($1.25-10/M), mini ($0.25-2/M), nano ($0.05-0.40/M).
• Claude Opus 4.1 & Sonnet 4.5 (Anthropic): Safety-focused, excellent for code and agentic tasks. Premium pricing (~$3-75/M tokens).
• Gemini 2.5 Pro (Google): Fully multimodal (text/image/audio/video), "thinking mode" for reasoning. Tightly integrated with Google's ecosystem.
• Qwen3 / Qwen2.5-VL (Alibaba): Open-licensed (Apache 2.0), competitive quality, multimodal, 128K+ context.
• Mistral Magistral (Mistral AI): Open weights, reasoning-capable, edge-deployable.

Key Trends

1. Agentic workflows - Models can now chain reasoning, call APIs, browse, execute code, maintain memory
2. Longer contexts - 200K-400K tokens becoming standard
3. Multimodal everywhere - Text+image+audio+video in single models
4. Safety & alignment - Red-teaming, guardrails, auditability now standard
5. Open weight revival - More customizable, fine-tunable, on-prem options
6. Cost commoditization - Even frontier models are dropping prices

Why Cloudflare AI Still Wins

Despite these advances, Cloudflare AI remains the most economical choice for 90% of use cases:

• ✅ Chat applications - Llama 3.3 70B matches GPT-5 quality for conversation
• ✅ Customer support - Fast responses, low latency, fraction of the cost
• ✅ Content generation - Creative writing, summaries, translations
• ✅ Code assistance - DeepSeek Coder rivals specialized models
• ✅ Prototyping & MVPs - Ship fast without budget concerns

When to use premium models:

• Multi-step reasoning over 100K+ tokens
• Mission-critical medical/legal analysis
• Advanced multimodal tasks (video understanding)
• Cutting-edge research requiring latest capabilities

For everything else? copilotedge-ai gives you 95% of the capability at 1% of the cost.

API Reference

copilotedge(options?)

Creates Next.js route handlers for CopilotKit.

Options:

• model - Model preset or name (default: 'FAST')
• telemetry - Enable telemetry (boolean or config object)
• temperature - Model temperature 0-1 (default: 0.7)
• maxTokens - Max response tokens (default: 1000)
• systemPrompt - System instructions
• debug - Enable debug logging
• accountId - Override Cloudflare account ID
• apiToken - Override Cloudflare API token

Returns:

• POST - Next.js POST route handler
• OPTIONS - Next.js OPTIONS route handler for CORS

presets

Pre-configured setups for common use cases:

• presets.chat() - Optimized for chat applications
• presets.code() - Optimized for code generation
• presets.creative() - Optimized for creative writing
• presets.support() - Optimized for customer support

Resources (work in progress)

ChatCloudflareWorkersAI

https://js.langchain.com/docs/integrations/chat/cloudflare_workersai/

Workers AI allows you to run machine learning models, on the Cloudflare network, from your own code.

This will help you getting started with Cloudflare Workers AI chat models. For detailed documentation of all ChatCloudflareWorkersAI features and configurations head to the API reference.

https://www.npmjs.com/package/@langchain/cloudflare

https://js.langchain.com/docs/concepts/chat_models/