Make your product catalog visible to AI shopping agents. This template serves a dynamic /llms.txt endpoint that uses Workers AI to transform raw product specs into natural language descriptions agents can reason with.

How it works: Connect your Shopify store (or use the included sample catalog), and the Worker enriches each product's technical specifications into agent-friendly summaries, use-case tags, and buyer highlights. The enriched catalog is cached in KV and served as a structured /llms.txt endpoint that any AI agent can consume in a single request. A small React SPA is bundled alongside the API at /, so a human visiting the domain sees the raw merchant data and the enriched output side-by-side.

Key features:

• Dynamic /llms.txt and /llms-full.txt endpoints following the llms.txt spec
• AI enrichment powered by Workers AI (Gemma 4) — turns "DIN 0.75-3.0" into "bindings release easily for toddler safety"
• KV-backed caching with configurable TTL — no re-enrichment on cold starts
• Shopify integration via public /products.json API
• Configurable merchant vertical to tailor AI descriptions per industry
• JSON API for programmatic access (/api/products, /api/products/:slug, /api/raw-catalog)
• Built-in React + Vite landing page that visualizes the raw → enriched transformation live against your own catalog

• Merchants who want AI agents to recommend their products. If you sell on Shopify and want ChatGPT, Perplexity, or other AI shopping agents to understand your catalog, this gives them a structured endpoint to consume.
• Developers building AI shopping experiences. If you're building an agent that recommends products, this is the merchant-side counterpart — a standardized, enriched product feed your agent can consume in one request.
• Platform teams exploring agentic commerce. If you're evaluating how to make product data agent-readable across a portfolio of stores, this template is a working starting point.

Outside of this repo, you can start a new project with this template using C3 (the create-cloudflare CLI):

npm create cloudflare@latest -- --template=cloudflare/templates/commerce-llms-txt-template

The template ships with a sample catalog so you can see it working immediately after deploy. Follow these steps to get live:

1. Install dependencies:

   npm install

2. Create the KV namespace used to cache enriched products:

   npx wrangler kv namespace create ENRICHMENT_CACHE

   Copy the namespace ID from the output and paste it into wrangler.json in place of the example namespace ID.

   If you deploy via the "Deploy to Cloudflare" button at the top of this README, the KV namespace is provisioned for you automatically — you can skip this step.

3. Deploy:

   npx wrangler deploy

Your /llms.txt endpoint is now live at https://commerce-llms-txt-template.<your-subdomain>.workers.dev/llms.txt. It serves the sample catalog with AI-enriched descriptions.

Configure your store in wrangler.json — set MERCHANT_NAME, MERCHANT_DESCRIPTION, and SHOPIFY_STORE_DOMAIN:

{
	"vars": {
		"MERCHANT_NAME": "Your Store Name",
		"MERCHANT_DESCRIPTION": "A short description for AI agents",
		"SHOPIFY_STORE_DOMAIN": "your-store.myshopify.com",
		"MERCHANT_VERTICAL": "outdoor gear"
	}
}

Then redeploy with npx wrangler deploy.

Once deployed, your Worker serves an agent-readable product catalog. Here's how to put it to use:

Point your domain at it. Add a Custom Domain or Route so /llms.txt is served from your store's actual domain (e.g., yourstore.com/llms.txt). AI agents and crawlers will discover it the same way they find robots.txt.

Test it with an agent. Try asking ChatGPT or another AI assistant: "What products does [your store URL] sell?" If the agent supports web browsing, it can fetch your /llms.txt and reason about your catalog directly.

Monitor usage. Use npx wrangler tail or the Workers dashboard to see requests to your /llms.txt endpoint and which agents are consuming it.

Set these in the vars section of wrangler.json:

For password-protected Shopify stores:

npx wrangler secret put SHOPIFY_STORE_PASSWORD

npm run dev      # Vite dev server + Miniflare for the Worker (HMR for the SPA)
npm run build    # Build the SPA + bundle the Worker (via @cloudflare/vite-plugin)
npm run preview  # Build, then run the production bundle locally
npm run test     # Run vitest tests
npm run check    # Type check + build + dry-run deploy

Under the hood the template uses @cloudflare/vite-plugin: a single vite build produces both the static SPA (output to dist/client/, served as Workers Static Assets) and the Worker bundle (uploaded by Wrangler). The Worker only handles routes it explicitly defines (/llms.txt, /llms-full.txt, /api/*). Any other request falls through to the SPA — that's what assets.not_found_handling: "single-page-application" in wrangler.json does.

When running locally without a Shopify store configured, the Worker serves a sample catalog of children's ski gear with hand-written enrichments. This lets you see the full /llms.txt output and the SPA without needing a live store or Workers AI connection. Miniflare provisions a local KV namespace automatically, so no extra setup is needed for npm run dev or npm run test.

On the first request (or after the cache expires), the Worker:

1. Fetches your product catalog from Shopify (or uses the sample catalog)
2. Sends each product to Workers AI for enrichment (batched 5 at a time)
3. Stores the enriched catalog in KV
4. Serves the cached result for subsequent requests

Cache layers: CDN caches the response for 5 minutes (Cache-Control: public, max-age=300). KV caches the enriched catalog for 1 hour by default (ENRICHMENT_CACHE_TTL). This means inventory changes can take up to ~65 minutes to propagate. Lower the TTL values if you need faster updates.

First-request latency: The first request after a cache miss will be slow — Workers AI needs to enrich every product. For a 50-product catalog, expect 5-15 seconds on the first request. Subsequent requests are fast (KV read + response).

• Public API only. This template uses Shopify's public /products.json endpoint, which doesn't require an API key. It works with any Shopify store that hasn't disabled this endpoint.
• Inventory is approximate. The public API exposes variant-level availability (in stock / out of stock) but not exact inventory quantities. The stockCount shown in /llms.txt is the number of available variants, not units on hand. For exact inventory, you'd need to integrate the Shopify Admin API.
• Password-protected stores are supported via the SHOPIFY_STORE_PASSWORD secret.
• Large catalogs are paginated automatically (up to 5,000 products). Be aware that very large catalogs will trigger more Workers AI calls on cache miss and may take longer on the first request.

• Workers AI cost on cache miss. Each product requires one AI inference call for enrichment. A 250-product store triggers 250 calls (in batches of 5) on the first request. These are billed as standard Workers AI inference requests. Once cached, no AI calls are made until the cache expires.
• No webhook-based cache invalidation. The cache expires on a timer (default: 1 hour). If you need instant updates when products change in Shopify, you'd need to add a webhook handler that clears the KV cache — not included in this template.
• Single-store only. The template serves one Shopify store per Worker. For multi-store setups, deploy one Worker per store or extend the routing logic.