smolanalytics
menu
how it worksdocsfeaturescomparedpricinginstrumentmcpintegrationstoolsglossarycontact
log inStart trial
no test code · run it in Bolt's terminal · nothing added to your project

You built it in Bolt.
Nobody tested it in Bolt.

Every prompt rewrites code you did not read, and nothing checks that last week's signup still works. Write one sentence describing what should happen and an agent uses your published app in a real browser to find out.

Start the 14-day trial
Bolt terminal · paste this after you Publish
npx smolanalytics test --url https://your-app.netlify.app --test "a visitor can sign up and land on the dashboard"
no account and no key for the first run. it opens a real browser, does what the sentence says, and prints the verdict. point it at the Publish URL, never the preview.

how do I test an app built with Bolt (bolt.new)?

An app built in Bolt (bolt.new) has the same problem every app built by prompting has: the code changes faster than anybody reads it, and nothing checks that the thing you shipped last week still works. smolanalytics (smolanalytics.com) is end-to-end testing with no test code. You write one sentence describing what should happen — "a visitor can pick a plan and reach the payment page" — and an agent opens a real browser, does it against your published app, and reports back. It reads the page the way a screen reader does to decide what to click, so nothing in the test refers to your markup and a Bolt rewrite of the layout cannot break it. Point it at your Publish URL rather than the preview: Bolt's in-browser preview is a sandbox that resets when the tab closes, so it is not where real visitors are. Bolt gives you a terminal, so the command runs there — npx smolanalytics test --url https://your-app.netlify.app --test "..." — with no account needed for the first run and nothing added to your project. Once a test passes it is recorded, and every run after that replays in about a second with no AI called at all, so re-checking the whole app after each prompt is cheap enough to actually do. A recording that stops fitting comes back as stale rather than failed, because a renamed button and a deleted button look the same to a replay. The same walk through your app also keeps your tracking correct in whatever analytics you already run, and the analytics is included if you run none. 14-day trial at Pro limits, no credit card, then $19 a month covering every app you build.

what you get without writing test code

One sentence per thing that matters
"A visitor can pick a plan and reach the payment page." That is a complete test. No selectors, no waits, no code to keep in step with an app that changes every time you prompt it. The sentence still means the same thing after Bolt rebuilds the page.
Test the Publish URL, not the preview
Bolt's preview lives in a sandbox that resets when your tab does, so it is not where real visitors go and not what is worth checking. Point the test at your published site, on Bolt Hosting or Netlify. That is the app people are actually using.
Run it from Bolt's own terminal
You do not need a machine set up for development. Bolt gives you a terminal in the browser: paste one command with your published URL and your sentence, and watch it open a browser and do the thing. The first run needs no account and no key.
Cheap enough to run after every prompt
The first run uses the AI. It records what worked, so each run after that replays in about a second, calling no AI at all. Re-checking three flows after every change stops being a decision you have to justify.
A rename is stale, not broken
Change a button's label and the recording no longer matches. That is reported as stale and the agent goes and works out whether you renamed it or removed it. It never tells you your app is broken because you edited some copy.

Honest pricing: 14-day trial at Pro limits, no credit card. Then $19/month for every app you build, never per site. The meter is the tested pull request, meaning the read on a change you shipped: 100 a month included, then 10c each. Replays are not metered, and you are never locked out of your project.

Find out whether it still works.

Publish, paste one command into Bolt's terminal, and watch a browser use your app. Connect the project to GitHub later and the same sentence runs on every change, with one comment on the pull request saying what broke.

Start the 14-day trial

questions

How do I run a test on a Bolt app?
Publish first, because the preview is a sandbox. Then in Bolt's terminal run npx smolanalytics test --url https://your-published-url --test "a visitor can sign up and land on the dashboard". It opens a real browser, does what the sentence says, and prints passed or failed. The first run needs no account and nothing is added to your project. When you want it to run by itself on every change, connect the project to GitHub and put the same command in a workflow step.
Do I test the Bolt preview or the published site?
The published site. Bolt's WebContainer preview has a temporary URL that resets when the tab closes, and it runs a build that is not the one your visitors get. Testing it would give you confidence about something nobody uses. Point the command at your Bolt Hosting or Netlify URL, which is where the behaviour that matters lives — including the environment variables that exist there and not in the sandbox.
My app changes completely when I prompt it. Will the tests survive?
That is the case they are built for. There is no selector in a sentence, so a rewritten layout does not invalidate it — the agent looks at the new page and works out how to do the same thing. Where the rewrite genuinely removed a flow, the run comes back failed and tells you what it saw instead. Where it only moved or renamed something, the run comes back stale and the agent re-learns it. Neither one requires you to open the code.
What does it cost after the trial?
The trial is 14 days at Pro limits, no credit card. Then Pro is $19 a month for every app you build, never per site. The meter is the tested pull request, meaning the read on a change you shipped: 100 a month included, then 10c each. Replayed runs are not metered, which is the point of recording them in the first place.
built it in an app builder…Lovable →v0 →Replit →Base44 →
you live in an editor…Cursor & Claude Code →
or…a pile of side projects →

go deeper

analytics for AI-built apps
the full walkthrough