Roadmap
Where Meerkat is heading, and which parts of it you can use today. Nothing on the second list exists yet, and there are no dates against it.
Last updated 18 September 2026
The direction
A usability test is something a team does before a launch, remembers for a week, and then forgets. We think the more useful shape is a product that keeps trying your flows and tells you when one of them gets worse — closer to uptime monitoring than to a research project. You would not check by hand whether your site is up.
Half of that is already true: every run compares itself with the previous run of the same flow, and a problem you marked fixed is flagged as a regression when it comes back. The other half — running without being asked, and telling you only when something changed — is not built.
Here now
- Runs you start
- AI personas try a task in a real browser and come back with what got in their way, worst first, with screenshots and a suggested change. How a run works.
- Findings that persist
- Mark a problem triaged, known, fixed or dismissed. Later runs recognise it, and say when something marked fixed has come back.
- Your own personas
- Describe a real customer in plain words and Meerkat builds a persona with the same traits the built-in ones have, for everyone in your workspace to run.
- Signed-in flows
- Save a login once and every persona of every later run starts already inside the product.
- An API, a CLI and webhooks
- Start a run from CI after a deploy, fail the build on a finding that gets in the way, and have the result posted to your own server. The API.
Next
These are the things we want to build, in the order we would build them. None of them exists yet, so do not buy Meerkat for them.
- A run on every deploy
- Connect a repository and have the flows you care about run themselves when you ship, instead of when somebody remembers. The API and CLI already let you wire this up by hand; this is about it being one setting rather than a job you maintain.
- Alerts only when something new breaks
- A run that finds the same three things as last time should be silent. The useful message is the one that says a flow got worse, or that a problem you fixed is back. Quietness is the feature.
- A trend line per flow
- One picture per flow, over time: where people get stuck, whether it is getting better, and which change moved it. Every run already records this; nothing reads it back yet.
- Confidence you can see
- Every finding should say how strongly it is believed and why — whether a screenshot confirms it, and whether more than one persona hit it. Our own study found that some of what the personas raised came from our test setup rather than from the product, and a score is only worth showing once it can tell those apart. The study.
How this page works
No dates, because we would only be guessing at them. Something moves from the second list to the first when you can use it, and the day it does it appears in the changelog. If something here matters to you, or something you need is on neither list, tell us and it will count.