Autonomous landing-page experiment

Tiny apps from one HTML file.

Deploy a single HTML artifact as a real app with install, analytics, shared state, cron-fed data, push, and agent-driven iteration.

Start free See MCP tools

Variant A: artifact-first framing. Tests whether “one HTML file” is the clearest wedge.

Ship a living micro-app in five minutes.

Turn HTML from Claude, ChatGPT, Lovable, or your own editor into a living app with one URL, one deploy, and one optimization loop.

Deploy your first app Import HTML

Variant B: runtime-first framing. Tests whether “living micro-app” produces more intent.

What the loop sees

Birta tags visits, screens, CTA clicks, forms, goals, and experiment variants. The agent reads those numbers daily and only rotates one challenger at a time.

Goal

signup

Experiment

homepage-loop-1

Variants

A / B

Method

sticky split

Tagging pattern

Use data-pc-variant, data-pc-goal, and data-pc-id on the important branches and CTAs.

Daily agent rule

No micro-tweaks. Only rotate a new challenger when the existing round has enough traffic to say something useful.

Three steps. One deployable loop.

This page is meant to be shipped as-is, then edited in place by an agent. Each section is deliberate enough to support meaningful challengers later.

01 · bring HTML

Start from the artifact.

Birta does not need a repo or a framework. Bring one HTML file, import it, and give it a real URL.

02 · make it live

Add data, memory, and push.

Add first-party analytics, shared or per-visitor state, cron-fed data, and installable PWA behavior without leaving the HTML model.

03 · let it improve

Close the loop with an agent.

Birta exposes the metrics back through MCP so the same agent that deployed the app can refine it over time.

What people actually build.

Same runtime, different use cases: landing pages, slides, trackers, weather surfaces, dashboards, tiny chats, and shared walls.

“I just want a living one-page app with one clean URL, not another full stack.”

Solo founder

“The fastest path is Claude → HTML → deploy → analytics → another pass.”

Agent workflow

“The interesting part is not generating it. The interesting part is that it keeps running after launch.”

Product loop

See the science in public.

The experiment log, current hypothesis, and scoreboard can live on separate public Birta pages. That makes the optimization loop visible before the final result is.

Open methodology page Open scoreboard Start the experiment