This is an experimental Worker. Use it as a starting point for your own projects.
Features
- Fetch and parse any public webpage
- Extract title and body text from HTML
- Generate 2-4 sentence summaries using Llama 3.1
- Sub-second response times on the edge
- No external API keys required
API Reference
GET /summary
Generate an AI summary of any webpage.
The webpage URL to summarize. Must use http:// or https:// scheme.
Response
The extracted page title from the HTML <title> tag.
A 2-4 sentence AI-generated summary of the page content.
Example Request
curl "https://your-worker.workers.dev/summary?url=https://www.cloudflare.com"
Example Response
{
"title": "Cloudflare - The Web Performance & Security Company",
"summary": "Cloudflare provides a content delivery network and DDoS protection services. The platform offers security, performance, and reliability solutions for websites and applications. It includes features like CDN, DNS, and web application firewall to protect and accelerate online properties."
}
Error Responses
Human-readable error message.
Machine-readable error code.
{
"error": "Missing or invalid query parameter: url",
"code": "INVALID_URL"
}
{
"error": "Failed to fetch or summarize",
"code": "FETCH_OR_AI_ERROR"
}
Implementation Details
AI Model Configuration
The worker uses Workers AI with the following configuration:
// From src/constants/defaults.ts
export const AI_MODEL = "@cf/meta/llama-3.1-8b-instruct-fast";
export const MAX_SUMMARY_TOKENS = 4096;
export const MAX_HTML_BYTES = 128 * 1024; // 128KB limit
export const FETCH_TIMEOUT_MS = 15_000; // 15 second timeout
Request Flow
The summarization process follows these steps:
Validate URL
Ensures the URL uses http/https scheme and is properly formatted:const url = validateUrl(c.req.query("url"));
if (!url) return jsonError(c, "Missing or invalid query parameter: url", "INVALID_URL");
Fetch HTML
Retrieves the webpage with a 15-second timeout and 128KB size limit:const html = await fetchHtml(url);
Extract Content
Parses the HTML to extract title and body text:const title = getTitle(html);
const bodyText = getBodyText(html, 6000); // Up to 6000 chars
Generate Summary
Sends content to Workers AI for summarization:const summary = await summarizeWithAi(c.env, bodyText, title);
- Generate exactly 2-4 concise sentences
- Return only the summary (no preamble or labels)
- Focus on key information from the page
Clean Response
Removes any boilerplate phrases the model may add:// Strips patterns like "Here is a summary:" or "Let me know if..."
return extractSummaryOnly(response);
Core Implementation
Here’s the main route handler from src/routes/summary.ts:
app.get("/summary", async (c) => {
const url = validateUrl(c.req.query("url"));
if (!url) return jsonError(c, "Missing or invalid query parameter: url", "INVALID_URL");
try {
const html = await fetchHtml(url);
const title = getTitle(html);
const bodyText = getBodyText(html, 6000);
if (!bodyText.trim()) {
return jsonSuccess<SummaryResponse>(c, { title, summary: "No extractable text content." });
}
const summary = await summarizeWithAi(c.env, bodyText, title);
const response: SummaryResponse = { title, summary };
return jsonSuccess(c, response);
} catch (e) {
const message = e instanceof Error ? e.message : "Failed to fetch or summarize";
return jsonError(c, message, "FETCH_OR_AI_ERROR", 502);
}
});
Setup & Deployment
Prerequisites
Local Development
Clone and install dependencies
git clone https://github.com/shrinathsnayak/cloudflare-experiments
cd cloudflare-experiments/experiments/ai-website-summary
npm install
Start the development server
This starts Wrangler in dev mode with Workers AI bindings. Test the endpoint
curl "http://localhost:8787/summary?url=https://www.cloudflare.com"
Deploy to Production
Authenticate with Cloudflare
Deploy the Worker
This publishes your Worker to *.workers.dev or your custom domain. Test the production endpoint
curl "https://ai-website-summary.YOUR_SUBDOMAIN.workers.dev/summary?url=https://blog.cloudflare.com"
One-Click Deploy
Click the button above to deploy this Worker directly to your Cloudflare account. You can fork the repository and update the URL to deploy from your own fork.
Configuration
The Worker automatically binds to Workers AI. The wrangler.toml configuration includes:
name = "ai-website-summary"
main = "src/index.ts"
compatibility_date = "2024-01-01"
[ai]
binding = "AI"
Dependencies
{
"dependencies": {
"hono": "^4.6.12"
},
"devDependencies": {
"@cloudflare/workers-types": "^4.20241127.0",
"typescript": "^5.7.2",
"wrangler": "^4"
}
}
Cloudflare Features Used
- Workers - Serverless execution environment
- Workers AI - Run LLMs on the edge with the
AI binding
- Fetch API - HTTP client for fetching webpages
Use Cases
- Content Curation - Automatically generate summaries for bookmarking tools
- Research Assistants - Quickly preview webpage content before visiting
- News Aggregators - Create summaries for article feeds
- Browser Extensions - Add “summarize this page” functionality
- Slack/Discord Bots - Share link summaries in team chats
Limitations
- JavaScript-heavy sites: Only static HTML is fetched; JavaScript-rendered content is not executed
- Rate limits: Workers AI has usage limits based on your plan
- Fetch timeout: Requests that take longer than 15 seconds will fail
- Size limit: HTML content is capped at 128KB; body text is truncated to 6000 characters
Next Steps