Definition

Cloudflare Workers is a serverless computing platform that executes code at Cloudflare’s network edge—over 310 cities in 120+ countries—using V8 isolates rather than containers or virtual machines as the isolation boundary. Developers deploy JavaScript, TypeScript, Rust (compiled to WebAssembly), or other WASM-compatible languages, and the code executes within milliseconds of the end user, in a fresh isolate that exists only for the duration of the request.

Workers were introduced in 2017 and represented a fundamental departure from the container-based serverless model pioneered by AWS Lambda in 2014. Where Lambda uses Firecracker microVMs with cold starts measured in hundreds of milliseconds, Workers use V8 isolates with cold starts under 5 milliseconds and memory footprints as small as 1-5 MB per invocation.

Why It Matters

Cloudflare’s network handles over 57 million HTTP requests per second, making it one of the largest edge computing platforms in existence. Workers execute within 50 milliseconds of 95% of the world’s internet-connected population—a latency guarantee that is architecturally impossible with centralized cloud deployments.

The isolation model is the critical differentiator for privacy workloads. AWS Lambda maintains warm containers for up to 15 minutes between invocations, meaning the memory state of a previous request—including any sensitive data it processed—persists and is theoretically accessible to subsequent invocations or the host operating system. Workers isolates terminate after each request. No warm pool. No reuse. No persistent memory.

In 2024, Cloudflare reported that Workers processes over 10 billion requests per day, with a p99 latency under 50ms globally. The platform’s pricing model (100,000 requests per day free, then $0.50 per million requests) has made edge computing economically viable for applications that previously could not justify the cost of global distribution.

For AI workloads that process sensitive prompts, the combination of edge proximity (reduced network hops), isolate ephemerality (no persistent memory), and Cloudflare’s AI Gateway (unified proxy to multiple LLM providers) creates an infrastructure layer purpose-built for privacy-first inference.

How It Works

A Cloudflare Worker request follows this lifecycle:

  1. DNS resolution: The client resolves the Worker’s hostname. Cloudflare’s Anycast network routes the DNS query to the nearest point of presence (PoP).
  2. TLS termination: The connection terminates at the edge PoP, within the same city as the client in most cases.
  3. Isolate creation: A fresh V8 isolate is instantiated for the request. If the Worker’s code has been recently compiled at this PoP, the compiled bytecode is cached (not the runtime state), reducing startup time further.
  4. Execution: The Worker’s code executes within the isolate. It has access to the Fetch API, the Web Crypto API, Cloudflare KV (global key-value storage), Durable Objects (stateful coordination), R2 (object storage), and AI Gateway (LLM proxy).
  5. Response: The Worker streams or sends the response back to the client over the existing TLS connection.
  6. Isolate termination: The isolate is destroyed. Its heap is deallocated. No memory state persists.

Workers have no filesystem. There is no /tmp directory. There is no writable volume. Data exists in the V8 heap during execution and nowhere else on the host.

Stealth Cloud Relevance

Cloudflare Workers is the compute substrate of Stealth Cloud. Every API endpoint—authentication, session management, message processing, session destruction—runs as a Worker at the Cloudflare edge. The architecture leverages three properties of Workers that no other serverless platform provides simultaneously:

  1. Sub-5ms cold starts enable ephemeral infrastructure without performance penalties.
  2. No filesystem means there is physically no location on the host where data could be written to disk.
  3. Global edge deployment means user prompts travel to the nearest PoP, not to a centralized data center—reducing the number of network hops and therefore the number of systems that could intercept traffic.

Combined with Cloudflare AI Gateway (which proxies requests to OpenAI, Anthropic, and other providers from Cloudflare’s IP addresses rather than the user’s), Workers allow Ghost Chat to process AI inference with zero metadata leakage from the user to the LLM provider.

The Stealth Cloud Perspective

Cloudflare Workers is not just where Stealth Cloud runs—it is why Stealth Cloud’s privacy guarantees are enforceable at the infrastructure level. A serverless platform with no filesystem, no persistent memory, and no warm containers is a platform that structurally cannot retain what it processes. That is not a configuration choice. It is a property of the runtime itself.