smolanalytics vs Playwright.
Reliable web automation for testing, scripting and AI agents, from Microsoft, open source and free.
Playwright or smolanalytics?
Keep Playwright. It is free, it is the best browser automation library there is, and nothing here replaces it. Buy this instead for the flows nobody got around to covering and the ones that go red every time a button is renamed: there is no selector to update, because there is no test code.
| smolanalytics | Playwright | |
|---|---|---|
| Price | $19/month with 100 tested pull requests included, then 10c each. 14-day trial at those limits, no card. The agent runs on your own Anthropic key, billed to you and never resold. | Free and open source, Apache 2.0. You pay in the time somebody spends writing and repairing tests. |
| How a test is written | One English sentence in a markdown file. No selectors, no page objects, no fixtures, no code to maintain. | Code in TypeScript, Python, .NET or Java, with a codegen recorder. Playwright Test Agents can plan, generate and repair suites with your own model. |
| When the UI changes | The agent looks at the page again and works it out. A recording that no longer fits is reported as stale, not as a failure. | The selector is updated by a person, or by the healer agent, and the file is committed. |
| At runtime | A recording replays with no model call; the agent wakes only when the recording stops fitting. | Generated tests run as ordinary code with no model at runtime. |
| What you maintain | Sentences. There is no test code in this product. | A test suite, its selectors, its fixtures and its page objects. |
where Playwright is the better pick
Playwright wins on every axis a library is judged by: it is free, it is faster, it is more capable, it runs on four languages and four browsers, and Microsoft maintains it. Their Test Agents will plan, generate and repair a suite using your own model, at no cost beyond tokens. If your team keeps its suite green today, there is no argument here worth reading: keep doing that.
where smolanalytics is
This exists for the case where that is not true. A suite that goes red on a rename gets muted, and a muted suite is worse than none because it looks like coverage. When a test is a sentence there is no selector to break, so the failure you get is a failure about your product. We also run under the code we test rather than beside it: one comment on the pull request, a named suspect file, and five verdicts that never blur a stale recording into a bug report.
How it works walks the loop from the sentence to the comment on the pull request, and the docs have the flags, the suite format and the CI step.
Pro $19/mo, 100 tested pull requests included, after a 14-day trial at Pro limits with no card. full pricing, including overage →
One sentence, on your app, tonight.
Point it at a URL that is already running and describe what should work. The first run uses the agent; once it passes it is recorded, so every run after that replays with no model at all.