Cloudflare Storage & Compute Demo

~10–12 min · Five services, one Worker, one URL · "Italic quotes" = phrases, your words are better · ▶ = action

Before the call

Cloudflare dashboard logged in · Dustin CF-1 Lab account
Live demo app open in browser tab: https://cf-demo-app.dustinburke23nc.workers.dev/
Second tab open on the D1 console (see "First time?" below for how)
Open a terminal with wrangler tail ready to run (don't start it yet)
Warm everything up 5 min before: hit /vectorize/seed, /r2/seed, /hyperdrive/query, /benchmark once each — first request to a cold Worker is slower

First time? Read this once, then never again.

Three things you'll touch during this demo. Each takes 30 seconds to set up.

1. Opening the D1 Console (the SQL editor)

This is a web-based SQL editor built into Cloudflare. No installs needed.

Open https://dash.cloudflare.com/ff89748ca36e597848348d987fc25108/workers/d1
You'll see a list with one database. Click "demo-d1-database".
At the top of the page you'll see tabs. Click the "Console" tab.
A big text box appears. That's where you paste SQL.
Type SELECT * FROM Customers; and click Execute. You should see 5 rows below.

Now leave this tab open. During the demo you'll switch to it and paste new queries.

2. Running `wrangler tail` (the live log streamer)

This shows logs from your Worker in real-time. Used during the Queues demo to prove messages are flowing.

Open Terminal (Cmd+Space, type "Terminal")
Run these two commands:

cd ~/Library/CloudStorage/GoogleDrive-dburke@cloudflare.com/My\ Drive/cf-demo-app
wrangler tail

You'll see "Connected to cf-demo-app, waiting for logs..." When you click buttons on the demo page, log lines appear here. Leave this terminal window visible — you'll point at it during the demo.

3. The demo page itself

Just a URL. Buttons trigger API calls; results show as JSON.

Open https://cf-demo-app.dustinburke23nc.workers.dev/
You'll see five colored cards (D1, R2, Vectorize, Queues, Hyperdrive) with buttons on each.
Click any button — a new tab opens with the JSON response from that service.

That's it. Three browser tabs and one terminal window. You're ready.

0. Open (~1 min)

1. D1 — Serverless SQL (~2 min)

Switch to the D1 Console

Paste this query, click Execute

SELECT c.CompanyName, c.Country,
       COUNT(o.OrderId) AS orders,
       SUM(o.Amount) AS lifetime_value
FROM Customers c
LEFT JOIN Orders o ON c.CustomerId = o.CustomerId
GROUP BY c.CustomerId
ORDER BY lifetime_value DESC;

"Joins, aggregates, ordering — all the SQL you'd expect. Production-ready. Every row update propagates to read-replicas around the world automatically — no ops, no read-replica management."

2. R2 — Object Storage (~2.5 min)

Show it working (45 sec)

Let the customer upload (45 sec)

▶ Click the orange "Upload your own file →" button at the bottom of the gallery

→ A drag-and-drop page opens. Hand them control: "Drag any file from your desktop — image, PDF, anything under 100 MB."

"What just happened: your browser sent that file as multipart form data to the Worker, the Worker called env.BUCKET.put() — one line of code — and R2 returned a success. The whole upload took ~100ms, and the file is immediately accessible at a public URL. No CORS dance, no presigned URL generation, no SDK config. This is one of the most-asked-about S3 patterns turned into one line."

⏸ Now the punchline. Go back to the gallery and scroll to the yellow cost box.

The egress story (60 sec)

3. Vectorize — Vector Database (~2 min)

Set up the concept (30 sec)

Run it (45 sec)

"My query is 'fast database.' Notice: none of my documents contain those exact words. A traditional keyword search returns zero. Vectorize returns the top match: Hyperdrive at 0.75 similarity — 'Hyperdrive accelerates queries to your existing Postgres database.' Second: D1 at 0.65 — 'D1 is Cloudflare's serverless SQL database.' The system understood that 'fast' relates to 'accelerate' and 'database' relates to both products, with no keyword matching at all. That's the premise of semantic search and RAG."

Land the payoff (30 sec)

"This is how you build a search engine over your own knowledge base — your internal wiki, your product documentation, your customer support history. It's how you build a chatbot that actually knows your product instead of hallucinating about it. And the entire flow you just watched — embedding the query, searching the index, returning ranked results — ran in one Worker call with two Cloudflare service bindings. No Pinecone API key, no OpenAI embedding bill, no separate vector database to operate."

4. Queues — Async Messaging (~1.5 min)

Set up the concept (30 sec)

"Queues are the platform's answer to 'do work in the background.' Every real application needs this — when a user submits a form, you don't want them waiting while your code sends emails, processes uploads, calls slow third-party APIs, or runs AI inference. You write the message to a queue, return the response immediately, and a separate handler processes the work asynchronously. Cloudflare Queues are bundled with Workers — no separate service, no separate billing, no separate SDK."

Run it (45 sec)

Land the payoff (30 sec)

"The top lines are the producer — every form submission, returned to the user immediately. The bottom lines are the consumer — Cloudflare batched all four messages together and called my handler once. That batching is automatic, and so is retry logic and dead-letter queue handling if a message fails. One Worker file contains both the producer that writes to the queue and the consumer that processes from it. Compare to SQS or Kafka — separate service, separate billing, separate IAM, separate SDK. Here it's two lines of code: env.QUEUE.send() to produce, and an async queue() handler to consume."

5. Hyperdrive — Postgres Accelerator (~3 min — this is the closer)

"Hyperdrive doesn't replace your database. It accelerates it. We pool connections to your existing Postgres or MySQL — anywhere — at the Cloudflare edge. We cache read queries. We eliminate the seven round-trips it normally takes just to connect before you can send your first query. This Worker is connected to a Postgres database on Neon, in us-east-1 — let me show you both paths side by side."

The Race — Direct vs Hyperdrive

▶ Click Race: Direct vs Hyperdrive (the red button on the Hyperdrive card)

What the JSON response looks like (numbers vary):

{ "test": "Hyperdrive vs Direct PostgreSQL Connection", "iterations": 5, "direct": { "label": "Direct connection (no pooling, no caching)", "connectMs": 40, // SLOW: full handshake every time "queryTimingsMs": [3, 3, 3, 2, 3], "totalMs": 54 }, "hyperdrive": { "label": "Cloudflare Hyperdrive (pooled and cached)", "connectMs": 2, // FAST: pooled connection "queryTimingsMs": [39, 4, 1, 2, 1], "totalMs": 49 }, "summary": { "speedup": "5.8x faster with Hyperdrive", "connectionSavings": "38ms saved on connection setup", "cacheBenefit": "First query: 39ms, cached: 1ms" } }

"What you're seeing: the same query against the same database, but on the left a direct Postgres connection — what your code would do normally. On the right — through Hyperdrive."

→ Walk the screen top-to-bottom

Direct connect time: ~40ms (TCP + TLS + auth handshake)
Hyperdrive connect time: ~2ms (connection already pooled and ready)
First query: similar — both hit the database
Subsequent queries: Direct ~3ms · Hyperdrive ~1ms (cached)

▶ Click the button again — second run is the punchline

"Notice the bottom of the response — '5.8x faster with Hyperdrive'. And remember — this database is on the same continent. For a database in Europe being read from Asia, Hyperdrive turns 200ms into 5ms."

6. AI Gateway — Fallback Routing (~2 min · skip if no AI workload)

Set up the problem (30 sec)

Show the happy path (30 sec)

▶ Click Fallback Demo button on the landing page · click Happy path (primary works)

Now break it (60 sec — the wow)

The Payoff — One Worker, Six Services (~1 min)

Recap — six products, one Worker

D1 — structured data (real SQL, with joins)
R2 — files (S3-compatible, zero egress)
Vectorize — AI search (semantic, not keyword)
Queues — background jobs (producer + consumer in one Worker)
Hyperdrive — your existing database, but fast (no migration)
AI Gateway — automatic fallback when LLM providers fail

"All bound to one Worker. One URL. One bill. One dashboard. Cloudflare isn't a CDN with bolt-on developer products — it's a complete application platform that also happens to be the world's largest CDN."

Close (~1 min)

If something breaks during the demo

Pre-recorded Loom of the full flow on the desktop
Screenshots of every JSON response saved as backup
If Vectorize returns empty matches: click Seed Sample Docs, wait 20 seconds, try search again. New embeddings take a moment to be queryable.
If Hyperdrive query errors: the Neon DB may have auto-scaled to zero. Click it once to wake — second click works fine.
Recovery line: "Perfect time to show you the dashboard." Open the corresponding Cloudflare dashboard tab and walk the data model. A graceful recovery beats a flawless demo.

Cloudflare Storage & Compute Demo — Cheat Sheet

Before the call

First time? Read this once, then never again.

1. Opening the D1 Console (the SQL editor)

2. Running `wrangler tail` (the live log streamer)

3. The demo page itself

0. Open (~1 min)

1. D1 — Serverless SQL (~2 min)

Switch to the D1 Console

Paste this query, click Execute

2. R2 — Object Storage (~2.5 min)

Show it working (45 sec)

Let the customer upload (45 sec)

The egress story (60 sec)

3. Vectorize — Vector Database (~2 min)

Set up the concept (30 sec)

Run it (45 sec)

Land the payoff (30 sec)

4. Queues — Async Messaging (~1.5 min)

Set up the concept (30 sec)

Run it (45 sec)

Land the payoff (30 sec)

5. Hyperdrive — Postgres Accelerator (~3 min — this is the closer)

The Race — Direct vs Hyperdrive

6. AI Gateway — Fallback Routing (~2 min · skip if no AI workload)

Set up the problem (30 sec)

Show the happy path (30 sec)

Now break it (60 sec — the wow)

The Payoff — One Worker, Six Services (~1 min)

Recap — six products, one Worker

Close (~1 min)

If something breaks during the demo

Quick Answers — back-pocket

Cloudflare Storage & Compute Demo — Cheat Sheet

Before the call

First time? Read this once, then never again.

1. Opening the D1 Console (the SQL editor)

2. Running wrangler tail (the live log streamer)

3. The demo page itself

0. Open (~1 min)

1. D1 — Serverless SQL (~2 min)

Switch to the D1 Console

Paste this query, click Execute

2. R2 — Object Storage (~2.5 min)

Show it working (45 sec)

Let the customer upload (45 sec)

The egress story (60 sec)

3. Vectorize — Vector Database (~2 min)

Set up the concept (30 sec)

Run it (45 sec)

Land the payoff (30 sec)

4. Queues — Async Messaging (~1.5 min)

Set up the concept (30 sec)

Run it (45 sec)

Land the payoff (30 sec)

5. Hyperdrive — Postgres Accelerator (~3 min — this is the closer)

The Race — Direct vs Hyperdrive

6. AI Gateway — Fallback Routing (~2 min · skip if no AI workload)

Set up the problem (30 sec)

Show the happy path (30 sec)

Now break it (60 sec — the wow)

The Payoff — One Worker, Six Services (~1 min)

Recap — six products, one Worker

Close (~1 min)

If something breaks during the demo

Quick Answers — back-pocket

2. Running `wrangler tail` (the live log streamer)