How it works

Eight stages between your prompt and a working app.

VULK isn't a single LLM call. It's a deterministic pipeline that classifies, plans, routes, generates, validates, repairs, previews and deploys — with the right model picked for each stage and an autofixer loop when a step fails.

  1. 01

    Fast classifier

    Classify the prompt

    What surface does the user want? Web app, mobile (Flutter), Shopify theme, marketing site, internal tool? A small, fast classifier picks the right pipeline before generation starts — wrong surface is the most expensive failure mode.

  2. 02

    GPT-5.6 Terra Pro

    Plan the architecture

    A planner expands the prompt into a feature list, schema sketch, route map, model decisions and acceptance criteria. The plan is what the user sees scrolling first; nothing is generated before the plan is locked.

  3. 03

    Router (deterministic)

    Route models per phase

    Your selected model writes the code. Auto applies VULK’s model policy, while request understanding and architecture planning have their own dedicated roles. The catalogue identifies each model’s role.

  4. 04

    Generator (per-phase model)

    Generate the code

    VULK generates the files defined in the project plan, validates their contents and records accepted files. Changes stay within the planned project, and accepted work remains available for later edits and recovery.

  5. 05

    Validator (rules + small model)

    Validate against the build

    Type-check, dependency resolve, vite build, schema sanity. Static analysis runs synchronously — we won't show you a preview that doesn't compile. Failures route into the autofixer; successes go straight to preview.

  6. 06

    Repairer (heavy model)

    Autofix the failures

    When the validator finds an error, the autofixer reads the actual error output and surgically edits the failing files. It loops until green or until a hard cap — and reports honestly when it can't fix something instead of pretending it did.

  7. 07

    Boot the microVM preview

    A Firecracker microVM provisions in seconds: kernel, rootfs, dependencies, the project. Live preview is real, not a sandboxed iframe — your generated app runs the same way it will in production.

  8. 08

    Deploy where it lives

    Publish your project to a live URL or export the code to your own infrastructure. Flutter uses its own build service for web previews and Android APKs, with app-signing credentials managed inside VULK.

Routing

Which model runs which stage.

Routing is deterministic and changes when benchmarks change. Specific model identifiers shift weekly — the policy below is stable.

StageModel classWhy
ClassificationGPT-5.6 SolUnderstands the request before generation
PlanningGPT-5.6 Terra ProBuilds the project plan and acceptance criteria
Generation (web)Heavy code-tunedTightest output rules
Generation (mobile)Specialised Flutter-awareWidget-tree correctness > brevity
Generation (3D)GPT-6 Astra with AutoDemands correct geometry / shaders
Edit (surgical)Heavy code-tunedSmaller diff window, must keep file alive
RepairHeavy reasoningReads compile errors as input
SummariseSmall / fastPlan recap, completion notes
The repair loop

What happens when generation goes wrong.

Most AI app builders stream code, hand it back, and hope. VULK assumes the first generation will fail at least once and treats recovery as a first-class step.

  1. Static checks first. TypeScript, eslint-blocking rules, dependency resolution. Cheaper than running the build.
  2. Build the project. Build and validate the project with its actual toolchain. Flutter goes through Dart analysis and a compiled web build, with build failures returned to the recovery flow.
  3. Capture the error verbatim. The fixer reads the actual compiler output, not a paraphrased summary.
  4. Surgical edit. The repair flow uses the actual validation error to correct the affected files. Files that already passed validation are preserved.
  5. Loop until green or cap. File generation has a bounded retry budget. If a file still fails, VULK reports the diagnostic, keeps the accepted files and lets you resume the interrupted generation.

Want to feel it?

The fastest way to understand the pipeline is to watch it run. Start on a paid plan — full plan credits, cancel anytime.

Last reviewed: April 30, 2026

Your idea, built and live in minutes.

Build my app
VULK Support

Online

Hi! How can I help you today?

Popular topics

AI support • support.vulk.dev