Cloudflare AI Resources (موارد كلاودفلير AI)

This skill enables the AI assistant to interact with and build applications using Cloudflare Workers AI and AI Gateway effectively, remaining strictly within the Free Tier limits and utilizing optimal architecture for billing protection.

[!IMPORTANT] Reference Documentation URL: https://developers.cloudflare.com/ Use this URL exclusively as the official reference for any queries related to Cloudflare Workers AI, AI Gateway, or compatible models.

1. Free Tier Overview & Allocation (الخطة المجانية وحدودها)

Cloudflare offers a highly generous free tier for Workers AI that resets daily.

• Daily Allocation: All Cloudflare accounts receive 10,000 neurons per day for free.
• Daily Reset: The allocation resets daily at 00:00 UTC.
• Neurons Concept: A "Neuron" is Cloudflare's unit of measurement for AI computing. Different tasks and models consume different amounts of neurons:
  • Smaller text models (e.g., Llama-3-8B) consume fewer neurons per token.
  • Larger models or image generation models consume more neurons per run.

2. Billing Protection & Cost Control (آليات الحماية والحد من الفواتير)

To ensure zero charges on your Cloudflare account, follow these absolute protection directives:

A. Keep the Account on "Workers Free Plan"

If your account has no paid plan active (Workers Free Plan), Cloudflare will enforce a Hard Limit. Once the daily 10,000 neurons are exhausted, subsequent API requests will fail with an HTTP 429 Too Many Requests error instead of charging you.

B. Use Cloudflare AI Gateway

AI Gateway acts as a proxy sitting between your application and the AI models, providing control, analytics, and caching.

• Enable Caching: Enable caching in your AI Gateway dashboard. Identical requests will be served directly from the cache without hitting the models, saving your daily Neuron balance.
• Configure Spend Limits (Gateway level): If a credit card is attached to the account, configure Spend Limits inside AI Gateway.
  • Set a daily/weekly/monthly budget in dollars.
  • Set rules to automatically block requests if the budget is reached.
• Set Account-Level Spend Limits: Set an account-wide limit for loaded credits (Unified Billing) as a final backstop.

3. Direct Instructions & Integration Methods (تعليمات وطرق الاستخدام المباشر)

You can call Cloudflare Workers AI via three primary methods:

Method A: REST API (External Applications)

• Base URL: https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/
• Authentication Header: Authorization: Bearer {API_TOKEN}
• Content-Type: application/json

Method B: Cloudflare Workers (Serverless Bindings)

When developing inside a Cloudflare Worker:

// wrangler.toml
[ai]
binding = "AI"

// index.js (Worker)
export default {
  async fetch(request, env) {
    const response = await env.AI.run('@cf/meta/llama-3.1-8b-instruct', {
      prompt: "Hello World"
    });
    return new Response(JSON.stringify(response));
  }
}

4. Multitask AI Orchestration Templates (نماذج بناء أدوات ذكاء اصطناعي متعددة المهام)

Here are the optimal boilerplate templates for generating Text, Images, and Audio.

A. Text Generation (Llama 3.1)

• Model ID: @cf/meta/llama-3.1-8b-instruct

Node.js / JavaScript

async function generateText(accountId, apiToken, promptText) {
  const response = await fetch(
    `https://api.cloudflare.com/client/v4/accounts/${accountId}/ai/run/@cf/meta/llama-3.1-8b-instruct`,
    {
      method: "POST",
      headers: {
        "Authorization": `Bearer ${apiToken}`,
        "Content-Type": "application/json"
      },
      body: JSON.stringify({
        messages: [
          { role: "user", content: promptText }
        ]
      })
    }
  );
  const data = await response.json();
  return data.result.response;
}

B. Image Generation (Stable Diffusion XL)

• Model ID: @cf/stabilityai/stable-diffusion-xl-base-1.0

Node.js / JavaScript

async function generateImage(accountId, apiToken, imagePrompt) {
  const response = await fetch(
    `https://api.cloudflare.com/client/v4/accounts/${accountId}/ai/run/@cf/stabilityai/stable-diffusion-xl-base-1.0`,
    {
      method: "POST",
      headers: {
        "Authorization": `Bearer ${apiToken}`,
        "Content-Type": "application/json"
      },
      body: JSON.stringify({ prompt: imagePrompt })
    }
  );
  // Returns raw binary image data (PNG/JPEG)
  const imageBuffer = await response.arrayBuffer();
  return imageBuffer;
}

C. Text-To-Speech (MelloTTS / Deepgram Aura)

• Model ID: @cf/myshell-ai/melotts or @cf/deepgram/aura-1 (if integrated)

Python Template

import requests

def text_to_speech(account_id, api_token, text):
    url = f"https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/run/@cf/myshell-ai/melotts"
    headers = {
        "Authorization": f"Bearer {api_token}",
        "Content-Type": "application/json"
    }
    payload = {
        "text": text,
        "lang": "en"  # or direct language code supported
    }
    response = requests.post(url, headers=headers, json=payload)
    if response.status_code == 200:
        # Returns binary audio file (typically wav/mp3)
        return response.content
    else:
        raise Exception(f"Failed to generate speech: {response.text}")

D. Audio-to-Text (Transcription: Whisper)

• Model ID: @cf/openai/whisper

cURL

curl https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/@cf/openai/whisper \
  -X POST \
  -H "Authorization: Bearer {API_TOKEN}" \
  --data-binary "@audio.mp3"

5. Sequential Multitask Workflow (سير العمل المتتابع متعدد المهام)

For orchestrating a pipeline where a prompt generates a story, makes an image for the story, and reads the story aloud:

graph TD
    A[User Prompt] --> B(Generate Story: Llama 3.1)
    B --> C[Story Text Output]
    C --> D(Text-To-Speech: MelloTTS)
    C --> E(Extract Key Visuals)
    E --> F(Generate Cover Image: Stable Diffusion)
    D --> G[Audio Output File]
    F --> H[Image Output File]
    G & H --> I[Consolidated Multitask Response]

To run this pipeline efficiently under the free tier:

1. Pass the prompt to Llama 3.1 to get the narrative.
2. Summarize or construct a visual prompt from the story.
3. Trigger Stable Diffusion for the image.
4. Trigger MelloTTS with the narrative to get the audio.
5. Utilize AI Gateway Caching on all prompts to ensure that running the pipeline repeatedly during development does not drain the 10,000 daily neuron limit.