Happo News
Get an email when we publish a new article or ship something noteworthy. Prefer to read in your own feed reader? Subscribe via RSS.
Recent articles
How we cut Happo queue times from 138 seconds to under 20
For years we absorbed traffic spikes in our screenshot workers by over-provisioning Kubernetes nodes and paying for capacity that mostly sat idle. Switching to KEDA autoscaling on queue depth, pre-warming capacity from an anticipated-work signal, and dampening that signal so estimates could not blow up the fleet cut average queue wait during bursts from 138 seconds to under 20. Here is the pattern, which is reusable for any bursty workload.
Introducing Happo's MCP server: let your agent review your visual diffs
Happo's new MCP server lets your AI agent review visual diffs the way you wish you had time to: structurally, with a written rationale, and with flake caught along the way. Because the agent already has your code change in context, it knows what UI changes to expect. Happo brings the diffs.
Ten years of the Happo logo: what changed, what stayed, and why the details matter
After ten years and three versions, we shipped a new Happo logo. Here's the story behind it: the rushed 2016 original, the smiling hippo that stuck, and a quiet typographic detail in the new wordmark that almost nobody will notice — and why caring about that kind of detail is exactly how we think about building Happo.
How to grow your Happo coverage without losing control of your bill
Happo coverage often stalls when teams worry about snapshot bills growing with their component count. Happo's new `--only` flag renders only the stories affected by a PR — dropping our own builds by 40% — so you can keep adding coverage without the cost anxiety.
Build vs. buy: scaling snapshot testing with Happo
When a major social media platform's in-house Android snapshot testing hit scaling limits with hundreds of developers, Happo built a Gradle plugin bridge. It preserved their existing Roborazzi setup while offloading storage, diffing, and reporting — showing that build vs. buy isn't always binary.