Pre-launch · design partners wanted

Stay online when the next zero-day drops

Run two independently built versions of your app. We shift traffic on the load balancer you already run toward whichever stack is least vulnerable, right now.

Out of band · config-only access · never in your request path

The problem

Monoculture is a single point of failure

Every replica is the same

Your fleet runs one language, one TLS library, one kernel. One bug in any of them is a bug in all of them.

Attackers set the clock

Disclosure starts a race between exploit and patch. Today you either take the risk or take the outage.

Redundancy isn't diversity

Multi-region and N+1 handle hardware and zones. They don't help when the vulnerable code is everywhere.

Two ways in

Route around risk. Or let us build the second stack.

Already run two diverse stacks? Plug in the risk engine. Only have one? We'll build its twin.

Managed service · waitlist

Least-Vulnerable Load Balancing

An out-of-band risk engine that watches threat feeds, matches them to what each stack really runs, and adjusts weights on your existing cloud load balancer.

  • Config-only access, never in the traffic path
  • Capacity guard: never drains every stack
  • Human approval for hard cuts, full audit trail
  • GCP first; AWS, Azure, Cloudflare next
Engineering service · quote

Stack Duplication

We'll duplicate your application in another language of your choice, on a different framework, OS and CPU, and keep both honest with one shared conformance suite.

  • AI-assisted rebuild, reviewed by engineers
  • Shared conformance tests: same inputs, same outputs
  • Separate pipeline, dependencies and keys
  • You own all the code
How it works

A risk engine beside your load balancer, not in front of it

The engine reads public vulnerability data and your signed inventories, decides, and writes traffic weights through your load balancer's config API. If we go dark, your load balancer keeps its last config.

Detect

OSV, CISA KEV, EPSS, distro and FreeBSD advisories, VEX.

Match

Signed SBOM per deployed image, cross-checked by an independent scan.

Score & check

Exploited? Likely? Reachable? Then capacity and flap guards.

Reroute & report

Approve, shift weights, verify health, log every decision.

Full walkthrough →

Architecture: threat feeds and verified inventory feed the Least-Vulnerable Risk Engine, which adjusts weights on the customer's cloud load balancer out of band. Stack A healthy at 100%, Stack B with a critical CVE drained to 0%, stay-up mode on standby.
Architecture overview. Customer traffic in blue, out-of-band control in orange.
Design diversity

Running Node? We mirror it in Java.

Java? Go. Go? Rust. The twin shares no runtime, framework, TLS library or package registry with the original, so one CVE can't take both down.

Running Node.js?

We mirror it in Java.

Event loop and npm, meet the JVM and Maven Central.

Running Java?

We mirror it in Go.

JVM and Spring to static binaries and the Go standard library.

Running Go?

We mirror it in Rust.

Garbage collector and crypto/tls to ownership and rustls.

Running Python?

We mirror it in Go.

CPython, Django or FastAPI to compiled Go services.

Running Ruby?

We mirror it in Elixir.

Rails and Puma to Phoenix on the BEAM.

Running PHP?

We mirror it in Kotlin.

Laravel or Symfony to Ktor or Spring on the JVM.

Running C# / .NET?

We mirror it in Java.

Kestrel and NuGet to Netty and Maven.

Running Rust?

We mirror it in Go.

Already on Rust? The twin goes the other way.

Examples, not a menu. Pick any target, or let us recommend the one that shares the least with your current stack. We also vary the OS, CPU architecture, build pipeline and signing keys.

Safety by design

Built to fail safe, not fail open

A tool that moves production traffic has to be more careful than the threat it reacts to.

01

Never in your traffic path

We hold config-only access to your load balancer. No proxy, no agent, no new software in the request path.

02

Capacity guard

The engine will not drain a stack if the remaining one can't carry the load. It never drains every stack.

03

No flapping

Hysteresis and dwell timers stop a noisy feed from bouncing traffic back and forth.

04

Fail static

Stale or unsigned feeds freeze changes. If we go dark, your load balancer keeps its last known-good config.

05

Humans approve hard cuts

Full drains need on-call approval by default. You choose what's automatic.

06

Every decision logged

Feed entry, inventory match, score, guard checks, approver and result, exportable for audits.

Open data, verified inventory

We match real CVEs to what you actually run

Not the Server header, not what's in Git. A signed SBOM tied to each deployed image, cross-checked by an independent scan.

Threat feeds

OSV.devCISA KEVFIRST EPSSDistro advisoriesFreeBSD SAsSigned VEX

Inventory

Signed SBOM per deployIndependent scanTied to live imageReachability
FAQ

Questions engineers ask first

Do you sit in front of my traffic?

No. The risk engine runs out of band and only changes backend weights through your load balancer's config API. Requests never touch our infrastructure.

What happens if Doppelstack goes down?

Nothing changes. Your load balancer keeps serving with its last configuration. Stale or unsigned feeds freeze changes rather than guess.

I only have one stack. Is this for me?

That's what Stack Duplication is for: we build a second, independently implemented version of your app and a shared conformance suite to keep the two in lockstep.

Is Stack Duplication a code conversion service?

No. It is a complete functionality mirror. We rebuild what your app does as an independent implementation on a different stack, and prove equivalence with a shared conformance suite. A line-by-line translation would carry your bugs across; a mirror is designed not to. More in the FAQ.

Is this available today?

Not yet. We're pre-launch and working with a small group of design partners, starting with GCP Global Load Balancer. Join the waitlist and we'll be in touch.

Least-vulnerable load balancing · waitlist

Get early access

We are onboarding a small design-partner group, GCP Global Load Balancer first. One email when your slot opens; nothing else.

You are on the list. We will email you when a design-partner slot opens.

Round robin. Least connections.Least vulnerable.