Midscene

Your AI Operator for Web, Android, Automation & Testing

[Midscene @midscene_ai]

Not "Playwright bad." For the 1000×-a-day "regress your own product" loop, selector maintenance compounds.

Dogfooding bought us back hours. Now we feel honest recommending it.

repo: https://t.co/Pm9Xyjmzh1

[Midscene @midscene_ai]

We use Midscene to test Midscene Studio.

Sounds like a tongue twister. It's our actual regression suite.

[Midscene @midscene_ai]

Designers tweak paddings, swap icons, move panels — none of this script changes.

When this lived in Playwright, every UI tweak triggered selector review. After moving to Midscene that kind of PR disappeared.

[Xuan Huang · 黄玄 @Huxpro]

A few of us leading @LynxJS_org @Rspack_dev @Midscene_ai (all open-source, 10k+⭐) will be in Bay/SF early June.

Looking to meet teams working on AI-native UI engineering.

DM me if you'd like to connect.

[Rspack @rspack_dev]

🏖️🦀TanStack Start + Rsbuild is a win-win for both communities 🤝

Rspack users get a full-stack choice: 100% TanStack Start, with router, SSR, RSC, Server functions out of the box

TanStack users can adopt the Rspack toolchain for stable bundled dev and first-class MF support

[Midscene @midscene_ai]

Different tools for different problems.

If your job is "agent that does the thing once" → check Browser-Use.
If your job is "agent that does the same thing 1000× reliably" → try Midscene.

→ https://t.co/Pm9Xyjmzh1

[Midscene @midscene_ai]

Question we get a lot: how does Midscene differ from Browser-Use?

Both are open source. Both use vision. Both work for what they're built for.

Here's the honest contrast — not a takedown.

[Midscene @midscene_ai]

Where Midscene shines: long-running E2E suites that survive UI changes. Same script across web + native. Cache replay so re-runs cost ~$0 tokens.

Where we're weaker: not as agent-y for free-form web exploration. Browser-Use's planning loop is more aggressive.

[Midscene @midscene_ai]

You can mix. Stable form ids → keep Playwright selectors. Icon buttons / dynamic content → Midscene AI calls.

Most of our active users migrated this way, not a clean rewrite.

→ https://t.co/kgRBeWlZx9

[Midscene @midscene_ai]

Migrating from Playwright to Midscene isn't rip-and-replace.

It's a 10-line diff. Keep your test runner, your fixtures, your CI. Just swap the locators where the UI keeps moving on you.

[Midscene @midscene_ai]

After — drop-in PlaywrightAiFixture:

await ai('click login, enter [email protected], expect success toast')

Same Playwright runner. Same fixtures. Same CI green checks.

Vision model finds the elements regardless of class names.

[Midscene @midscene_ai]

→ https://t.co/Pm9Xyjmzh1 (13k stars, MIT)
→ https://t.co/qtVWis8sLy

Issues, PRs, Discord all active.

[Midscene @midscene_ai]

Last one for anyone who hasn't tried Midscene.js yet — here's what runs on the same script.

[Midscene @midscene_ai]

Best case we've seen: an IoT team covering their app on Android + iOS + HarmonyOS plus the web admin console with one suite.

Manual QA hours saved per week measured in headcount, not minutes.