SanityCheck launched this week with a proposition that targets every development team that has cut corners on QA because writing and maintaining test code takes time they do not have. The tool spins up AI agents that run through your web application in real browsers, clicking through the same workflows a human tester would, and delivers a report with a full replay of every action. You describe what you want tested in plain English. The agents figure out the rest.
The company is betting that the language barrier between "I want to make sure the checkout flow works" and "write a Cypress test for the checkout flow" has kept too many teams from testing at all. SanityCheck removes that barrier. You paste a URL, write a sentence describing the user journey you want verified, and the platform does the rest. No SDK, no CI configuration, no test framework to learn.
How the agents actually work
When you run a test, SanityCheck launches a real browser instance in the cloud. An AI agent takes control and navigates through your application exactly as a user would, clicking buttons, filling in forms, and verifying that the expected outcomes appear. You can watch the session live as it happens, or let it run in the background and review the results later.
Every action the agent takes is logged alongside console output, network requests, and errors. These are time-synced to a video replay of the session. If something breaks, you see precisely when it broke, what the agent was doing at the time, and what the browser console reported. The forensic replay eliminates the back-and-forth of reproducing a bug manually.
For teams that need to test under load, SanityCheck supports launching multiple concurrent browsers from a single trigger. Each agent runs through the same test independently, giving you a combination of regression testing and load testing without separate tooling. If ten agents all try to create an account simultaneously and one fails, you know exactly which instance broke and under what conditions.
Credential handling matters
One of the harder problems in automated testing is authentication. Most web applications require login, and passing real credentials to an automated agent raises obvious security concerns. SanityCheck encrypts test credentials and fills them through tool calls rather than exposing them to the model. The agent logs in without ever reading the password in plain text.
You can also set hard limits on every run. Allowed domains restrict where the agent can navigate. Step and time budgets prevent runaway sessions. A "never do" ruleset lets you tell the agent what it must not do, such as deleting data or sending real emails. These constraints are enforced by the platform on every execution, not by the agent's interpretation of your instructions.
Deploy hooks close the loop
SanityCheck integrates with GitHub, GitLab, and Vercel to trigger tests automatically when a deployment goes live. On GitHub, you install the app on a repository and select which tests run when a deployment is reported successful. No changes to your workflow files. On GitLab, you authorize the connection once and pick a project, with tests firing when a pipeline or deployment succeeds, filtered by branch or environment. Vercel users point a webhook at SanityCheck and every successful deploy, production by default, gets tested the moment it goes live.
This turns what is usually a manual or semi-automated process into something that runs without human intervention. Every time you push code, a team of agents is already clicking through the critical paths of your application. The results show up in the SanityCheck dashboard, giving you a pass or fail status before you even open your inbox.
The tagline the company uses captures the appeal directly: "Before I ship, make sure a normal user can still do the important things in my app, without me writing test code." For teams that have been shipping without comprehensive QA because the overhead of building and maintaining a test suite was too high, this is a lower barrier to entry than any existing framework.
What you get for free
The free tier includes 15 agent test runs per month. No credit card required, no test code to write, nothing to maintain. For small teams or solo developers shipping quickly, that gives you coverage on the most critical user flows without committing to a paid plan. If you need more runs or concurrent agents, paid tiers are available.
The pitch is straightforward: ship like you have a QA team. Whether SanityCheck can deliver on that at scale depends on how reliably the agents handle complex, multi-step workflows across different applications. But the approach addresses a real gap. Most teams know they should test more. Few have the resources to build and maintain the test suites that would let them do it. SanityCheck is trying to make the testing itself as easy as describing what a user should be able to do.
For developers who have been shipping code and hoping nothing broke, the tool is worth watching. For teams that have been meaning to add QA coverage but keep deprioritizing it, it might be the lowest-effort way to start.