In its first 27 days, Unkey used Polylane to review every pull request against live production across Vercel, Cloudflare, PlanetScale, ClickHouse and Kubernetes, and to fix the issues that slipped through anyway.
Connected to Polylane
The challenge
When your product is an API that other companies call on every request, a small mistake does not stay small. A renamed field or a changed request body can break a customer's integration the moment it deploys, and the first signal is usually a support ticket.
Unkey runs across five platforms: Vercel, Cloudflare, PlanetScale, ClickHouse and Kubernetes. Tests cover the code in a pull request, but they cannot see everything that depends on it: the dashboard calling the API, the services that run the old and new versions side by side during a deploy, the data paths that change shape underneath.
“Our API sits in the request path of every customer we have. A bad deploy on our side becomes an outage on theirs. Polylane reads every pull request against what is actually running in production, and it has caught things that our tests and our reviewers both missed. The first one was two hours after we signed up.”
Andreas Thomas Co-founder & CTO at Unkey Why Polylane
Polylane built a context graph of Unkey's stack across all five platforms, 66,747 cloud resources in its first 27 days, and connected it to Unkey's repositories, Sentry and Grafana. When a pull request opens, Polylane reviews the change against that graph: what calls this code, what runs it in production, and what happens during the deploy while both versions are live.
Two hours after Unkey signed up, Andreas opened a pull request that changed the request body schema of an API endpoint. Polylane flagged that one of the dashboard pages would crash until the API was fully deployed. It was the first of 10 risky changes Polylane stopped before merge in its first 27 days.
How Unkey uses it
Every pull request at Unkey gets a Polylane review alongside the human one: 385 in its first 27 days. Polylane posts its findings on the pull request with the evidence from production: the callers it found, the traffic they carry and the window in which they would break.
When something slips through, Polylane detects it in production, investigates across code, deploys and telemetry, and opens a fix that goes through Unkey's normal review and CI. Issues and fixes land in Slack and Linear, where the team already works.
The results
In Unkey's first 27 days, Polylane monitored 66,747 cloud resources, reviewed 385 pull requests and flagged 10 changes that would have broken something in production before they merged. It detected 498 anomalies, surfaced 63 issues worth investigating, and opened fixes with a median of 18 minutes from detection to a pull request ready for review.