Life of a Vercel request: Application-aware routing

Link to headingRoute-aware and application-aware infrastructure

Vercel uses your framework code to define infrastructure and how requests are handled at runtime. This tight integration gives the platform full visibility into your app’s structure: routes, layouts, rewrites, middleware, functions, static assets, and routing logic.

We refer to this as route-aware and application-aware infrastructure. That awareness powers the platform’s ability to optimize everything from build output to request resolution, fulfillment, and delivery.

Unlike a traditional standalone CDN, Vercel's gateway doesn’t need to be manually wired to backends or hardcoded to interpret intent. Routing is not bolted on—it's defined by your app and built into the deployment. This allows it to resolve requests across millions of domains and deployments in real time.

Link to headingAliasing domains on Vercel

Every domain—whether it’s your custom domain purchased through Vercel or a third-party, or a free *.vercel.app domain—serves as an external entry point from the internet.

Behind the scenes, these domains are aliases—logical mappings to a deployment within Vercel’s global routing.

Each deployment, shaped by the framework-defined build process, includes its own routing table, middleware, headers, and logic—providing full app context for request evaluation and WAF logic. When a domain is aliased to point to a deployment, Vercel automatically provisions and manages SSL/TLS certificates for HTTPS with no manual configuration, risk of expiration, or downtime.

Link to headingDe-aliasing and early stages of request fulfillment

When a request is made to your domain, it travels via Anycast routing to the nearest Point of Presence (PoP), passes through initial DDoS mitigations, and lands in a Vercel Region for initial Web Application Firewall (WAF) inspection.

The request is then enriched in real-time, through de-aliasing—the reverse process of mapping the domain to a specific deployment.

While traditional WAFs operate blindly, evaluating requests without understanding the app they’re protecting, Vercel’s WAF is made application-aware, as it's integrated with all routing metadata associated with a deployment. That means firewall rules can target specific parts of your application, not just raw path or cumbersome regex patterns.

At this point, and throughout the entire lifecycle of the request, Vercel's routing can return with an early response, serve from cache, or invoke compute. In a dynamic network, the destination isn’t fixed as the gateway decides where to send each request.

Link to headingA platform unlocked by application-aware routing

Vercel’s gateway provides automatic deployment, routing, and rollout behavior, thanks to the deployment-based approach (quick aliasing and de-aliasing) combined with an instant global propagation pipeline.

Let’s look at a few examples.

Link to headingInstant, global propagation for rollouts and rollbacks

When you alias a domain to a deployment, which could be an existing deployment we're updating or a brand new one created from a PR a few seconds ago, that alias becomes the active entry point for all traffic to that domain.

Switching that alias—mapping to an internal Vercel deployment—happens instantly. There’s no DNS propagation, no warmup time, and no manual steps. The gateway starts routing traffic to the new version immediately. This change is propagated globally so all users requesting your app will hit the latest deployment milliseconds after it goes live.

If a deployment or PR needs to be reverted, you can just as quickly re-alias the domain to the previous deployment with Instant Rollback—which remains untouched given its immutability—and traffic instantly directs to that deployment in a known good state. Importantly, this keeps the CDN cache intact, and is all done with no rebuild, no redeploy, no scale-down event. Just a global pointer update.

Link to headingIncremental rollouts

This also enables incremental rollouts of new deployments, often referred to as blue/green deployments. With all Vercel deployments, you can test a new version using preview deployments to run health checks, and only promote it to production once it’s ready—all while keeping the previous version one alias change away. Because aliasing is a pointer, you can modify routes to roll out new builds in phases, where there are multiple production environments live at the same time.

If there are issues during the incremental rollout of green (new, meant to be fully live deployment), an instant rollback returns the percentage of traffic back to 100% to blue.

Link to headingSkew Protection against version mismatches

Ensuring rollouts don’t affect users who are actively using your app can be tricky to manage. Backend logic can get out of sync with frontend code running in user's browsers.

Skew Protection helps prevent mismatches between frontends and backends by ensuring a user stays on the same deployment with aliasing pointing to specific deployments. It keeps track of which deployment every user of your app is accessing at any given point and intelligently keeps them in sync.

If you deploy a new frontend that depends on changes in the backend, you want to make sure that clients using the new UI aren’t sending requests to an outdated version of the infrastructure. With Vercel’s gateway, de-aliasing solves this. A request that includes a deployment identifier, via a first-party cookie, can bypass the alias entirely and be routed to the exact deployment that originally rendered it.

Link to headingA continuous, intelligent reverse proxy

The gateway isn’t just deployment-aware—it’s programmable, serving as an intelligent reverse proxy in front of your apps, facilitating almost all routing. With the integrated gateway, routing doesn’t happen once. It happens continuously.

Throughout the request's lifecycle, the gateway ushers the request through the steps laid out by framework code from the previously mentioned aliasing and de-aliasing and WAF inspection, to programmatic middleware execution and dynamic rewrites. The ushered request is continuously re-evaluated for either continued fulfillment, an early return, or with rewrites that go on yet another path to complete its fulfillment.

This is all derived from code. Developers shape routing at every stage—before, during, and after inspection—with the logic they've defined and versioned with every immutable snapshot of code. This continuous evaluation isn’t just passive—different decisions can be made on the fly for every request, like with middleware.

Link to headingProgrammatic routing functions

Middleware runs during routing, giving you full access to the request object (headers, cookies, pathname, and more). This logic executes close to your users and can short-circuit the request, mutate the path, inject headers, or return early responses. Middleware lets you, with very low latency, inject feature flags and A/B experiments, first-line authentication checks for quick returns, geo-based redirection, application-level bot filtering, and session-aware routing.

Many routes don’t require dynamic logic. They’re based on predictable patterns or static conditions like pathnames and headers. In these cases, Vercel’s routing layer supports versioned, framework-defined route maps that enable precise control without the need for runtime code execution. This is especially useful when multiple underlying apps serve different parts of the same domain.

Link to headingRewrites and redirects

Rewrites and redirects are first-class primitives. Rewrites route a request to a different origin or path—internal or external—without changing the URL users see. This is critical for building systems like microfrontends, where different parts of an app may live in separate deployments.

For example, you can rewrite /blog/[slug] to a headless CMS or /dashboard to an internal service. The routing layer ensures the client stays unaware of these boundaries, preserving a unified app experience. Rewrites can also target fully external URLs while keeping the original route structure intact.

This allows multi-tenant apps like Dub and Mintlify to use wildcard subdomains, which allow users to set custom domains easily. This points many different hosts to the same app (which may be backed by many apps and deployments) with centralized routing and minimal complexity.

In contrast, redirects update the browser URL and enforce client-side routing changes. They’re useful when migrating legacy paths, fixing SEO issues, or enforcing canonical URLs.

Both rewrites and redirects can be defined statically in your config or generated dynamically via middleware, and both are fully tied to the active deployment. There’s no global routing config that sits outside the lifecycle of your app. Everything is scoped to your code and versioned along with it.

Link to headingFramework-defined routing

The Vercel gateway is an intelligent routing system with full awareness of both provisioned infrastructure and every deployment. What’s needed is shaped by your app’s code, giving users the ability to control real-time routing with programmable logic.

As each framework handles routing differently. Vercel analyzes, infers, and then preserves the routing output of every build to ensure requests are handled according to the framework’s behavior and coded intent. It doesn’t just serve static assets or forward requests. It understands the structure and state of your app. It knows what deployment is active, how requests should be shaped, and what logic should run before a response is served. It gives you control without requiring infrastructure management, and flexibility without sacrificing predictability.

Routing is more than matching paths to handlers. On Vercel, it’s a dynamic process that spans deployments, domains, and logic—without drift or downtime. The gateway is where traffic is understood, where code paths are chosen, and where infrastructure becomes part of your app's runtime behavior.