How we push a kill switch to 312 edge nodes in 38 ms
A kill switch that takes a minute to arrive is a status page with extra steps. How our control plane fans a single write out to 312 edge nodes, and where the 38 ms actually goes.
Feature flags, guarded rollouts and kill switches that reach every edge node in 38 ms. Built for teams that deploy every day and sleep every night.
new-checkout-flow
One-page checkout with saved cards
edge-cache-v2
Stale-while-revalidate at the edge
search-reranker
Semantic reranking for site search
Rulesets stream to 312 edge nodes. Your code never waits on a network call to ask whether a flag is on.
User attributes never leave your servers. Driftwire sees flag keys and counts — nothing that identifies a person.
Any flag can be an experiment. Results arrive with the confidence interval already worked out.
An OpenFeature provider for every SDK, so the flag calls in your code are not a lock-in.
Platform
Every change ships dark, rolls out by percentage and stops itself when a guardrail trips. The deploy and the release are finally two different things.
Turn anything off, everywhere, before the alert finishes paging.
1% to 100% on a schedule, paused automatically when error rates move.
Segments, rules and per-user overrides, evaluated locally in every SDK.
Who changed which flag, when, and what it was before. Exportable.
Two of the teams who moved their releases behind Driftwire, and what changed.
4 sto roll back, from 40 min
“Rollback used to be a 40-minute pipeline. Now it is a toggle, and on-call is boring again — which is the highest praise I have.”
2,000flags migrated in a weekend
“We moved 2,000 homegrown flags in a weekend and deleted 312 of them the next month. Nobody noticed, which was the point.”
Developers
Install the SDK, wrap the change, and every decision after that is a click — not a deploy. Typed clients for the languages you already ship.
Rules live in memory in every SDK. A flag check is a function call, not a request.
Changes arrive over a long-lived connection, so a kill switch lands in milliseconds.
If Driftwire is unreachable, your code serves the defaults you set. Nothing breaks.
import { createClient } from '@driftwire-demo/sdk'
const driftwire = createClient({ sdkKey: process.env.DRIFTWIRE_SDK_KEY! })
export async function checkoutFor(user: User) {
const useV2 = await driftwire.isEnabled('new-checkout-flow', {
key: user.id,
plan: user.plan,
})
return useV2 ? renderCheckoutV2(user) : renderCheckout(user)
}<script setup lang="ts">
const { enabled } = useFlag('new-checkout-flow')
</script>
<template>
<CheckoutV2 v-if="enabled" />
<Checkout v-else />
</template>import express from 'express'
import { createClient } from '@driftwire-demo/sdk'
const driftwire = createClient({ sdkKey: process.env.DRIFTWIRE_SDK_KEY })
const app = express()
app.get('/checkout', async (req, res) => {
const on = await driftwire.isEnabled('new-checkout-flow', { key: req.user.id })
res.render(on ? 'checkout-v2' : 'checkout')
}){
"flags": {
"new-checkout-flow": {
"rollout": { "steps": ["1%", "5%", "25%", "100%"], "every": "4h" },
"guardrails": [{ "metric": "error_rate", "below": "1%" }]
}
}
}$ npm i @driftwire-demo/sdk
$ driftwire flags create new-checkout-flow --type boolean
$ driftwire rollout new-checkout-flow --env production \
--to 25% --guardrail "error_rate < 1%"
› 5% to 25% · propagating to 312 edge nodes
ok live everywhere in 38 msIntegrations
Flag changes post to your team chat, link to the pull request that shipped them, and show up in your traces. SDKs for every runtime you deploy.
“Guardrails turned rollouts from a meeting into a schedule. Releases advance overnight and stop themselves if something looks off.”

“We ship forty times a day with a team of nine. Driftwire is the reason that number does not scare our board.”

“The audit log settled an argument in our incident review in about eight seconds. That alone paid for it.”

A kill switch that takes a minute to arrive is a status page with extra steps. How our control plane fans a single write out to 312 edge nodes, and where the 38 ms actually goes.
A 25% rollout is not "one request in four". It is a stable answer to "is this user in?", identical on every server, that never flips a user back out. That is a hashing problem.
Lumenfold deploys 60 times a day to 4 million readers. Their rollback used to be a 40-minute redeploy. Now it is a toggle, and their on-call rota is boring again.
Free for five seats and 10,000 monthly users. Your first flag is about ten minutes away, with nothing to pay and nobody to talk to.