Desktop screenshots
The generated pages are rendered at desktop width and reviewed before you open them.
AI BUILDER
Describe your idea. Get a site that was already tested.
Hostingsurge AI Builder turns a prompt into a real project: plan, design, sitemap, content, code, a private version commit, an isolated build and a browser test run on desktop and mobile — before the preview reaches you.
Availability: the builder pipeline, the AI router and the isolated build runner are implemented and verified in development. Live model inference starts when the platform model provider is configured — until then a run reports not configured instead of inventing output.
Intent to planDesign systemSitemap and contentCodePrivate versionIsolated buildBrowser QARepair loopPreview
From one sentence to a tested preview.
The builder works through a fixed pipeline and shows concise progress and outcomes — never chain-of-thought.
- Intent extractionThe prompt is read for purpose, audience, language, sections and must-haves before anything is generated.
- Project planA concrete plan: pages, components, content structure and the technology that actually fits. Static output wherever the site can be static.
- Design systemType scale, spacing, colour and component rules derived from the brief instead of a template gallery.
- Sitemap and contentEvery page gets one purpose and real copy for your business. No placeholder text is allowed to survive into the output.
- Code generationThe result is a real project with source you own — not a locked page builder file you cannot leave.
- Private version commitEach run is saved as a version in the project private history. You work with versions, restore, compare and export; no Git knowledge is required.
- Isolated buildThe build runs with no network, capped memory, CPU and process limits, a read-only filesystem and a bounded wall-clock timeout.
- Browser QAThe built site is opened in a real browser on desktop and mobile: console and network errors, forms and critical controls, accessibility and basic SEO.
- Automatic repair loopFindings are fixed and re-tested inside a configurable, budget-aware repair budget — so you are not the first person to test your own site.
- Preview readyYou get a preview together with the real state of the run: what passed, what was repaired, what still needs your decision.
You are not the first QA tester.
A build only reaches your preview after the machine part of the review is done.
Mobile screenshots
The same pages are rendered at mobile width, so the layout that matters most is checked too.
Console and network
JavaScript console errors and failed network requests are collected and reported as findings.
Forms and controls
Critical controls and forms are exercised instead of being assumed to work.
Accessibility checks
Semantics, focus states and contrast are part of the review, not a later add-on.
Basic SEO checks
Titles, structure and the basics search engines read are validated together with the build.
Progress you can follow.
Runs report real state. A queued, building or failed run is never displayed as ready.
// live run progress Planning site Creating design Building pages Testing mobile Checking links Optimizing Ready
Build Credits, not token counting.
Credits are separate from infrastructure resource units, and one credit is not a fixed number of tokens. Work is estimated before a multi-step build, reserved, and settled against the real outcome afterwards.
| Typical action | Guided range |
|---|---|
| Change a button colour | 0.25 – 0.50 credits |
| Edit a component or a section of copy | 0.50 – 1.00 credits |
| Add a normal page | 1 – 2 credits |
| Build a landing page | 2 – 4 credits |
| Add a backend feature | 2 – 5 credits |
| Ship a larger feature | 3 – 8 credits |
| Build a new app | 5 – 15 credits |
Plans and credits
- Free5 one-time Build Credits after verified signup
- Pro — $25 / month100 monthly Build Credits
- Business — $50 / month100 monthly Build Credits plus teams, admin and policy controls
- Top-up$25 = 100 Build Credits
- EnterpriseCustom pooled terms
Guided ranges from the specification, not a published price list. Failed jobs release their reservation, and no task is sold below its configured cost floor.
Model routing you never have to manage.
Every request goes through the platform AI router rather than a model hardcoded into the product, so routing can change without a deployment.
- Task routes are separate: planning, code, fixes, design and vision review each get a route that fits them.
- Provider fallback, retries, timeouts, structured-output validation and tool-call validation on every call.
- Budget guards, a maximum agent iteration count and loop detection around multi-step runs.
- Token, cost, latency and success or failure are recorded per attempt instead of being summarised away.
- Private repository content is never sent to a model or provider that has not been approved for private source code.
- Free model routes are only eligible when admin policy and the project data classification both permit them.
Edit what you see, not the code.
AI chat is the primary way to change a project. Lightweight visual editing covers the everyday changes.
Edit text in placeReplace an imageModify linksAsk AI about the selected componentHide or show an elementReorder safe sectionsDuplicate a section
Visual edits are not a parallel world: they produce source-code changes and commits, so the project stays a real project. Visual editing is in development and lands after the core builder pipeline.
Where this stands. The builder pipeline, the AI router, the isolated build runner, the artifact store and the version history are implemented and verified in development, and a project view shows the latest run with its real state. Live inference, the full onboarding sequence and visual editing are still being wired. Nothing here claims a completed public launch, and no model output is fabricated when a provider is missing.
Questions, answered plainly.
What the builder does, what it needs and what happens when something fails.
Do I need a GitHub account?
No. Every managed project gets a private repository inside the platform source service, and the customer interface talks about versions, restore, compare and export. Developers can connect GitHub or GitLab later, but it is never required to build with AI.
Which model does it run on?
Requests are routed through the platform AI router. The default is a flash-class model for planning, coding and repair, with separately approved routes for hard reasoning and vision review. The model is configuration rather than business logic, so it can change without a code deployment.
What happens when a build or a test fails?
The run ends in a real failure state with an honest reason. The previously saved version is untouched, reserved credits are released instead of consumed, and the platform never reports a failed run as ready.
Can I take the result with me?
Yes. Source and versions are exportable at any time, and a bounded ZIP export of a saved version is part of the implemented source work. The platform is not a cage around your project.
Does my code get sent everywhere?
Before a model may ever see it, the route must be approved for the data class private source code. Free routes and unapproved providers are excluded from that path by policy.
Build it. Run it. Own it.
Start with a prompt and keep the source. The builder produces a real project you can export, move and reason about.