Architecture

How the engine works.

Gallop isn't one model doing everything. It's a pipeline of specialized agents, each constrained by schema, working in parallel to generate production WordPress content.

Read the docs

The Pipeline

Three agents. One page.

Each agent is a specialized model with a narrow scope and strict output constraints. They run in parallel, not sequentially — that's why generation takes seconds, not minutes.

01

Layout Agent

Determines page structure: section count, row configurations, column widths, and element types. Outputs a structural blueprint — a JSON tree of containers — constrained by the element schema. No content, no styling. Just structure.

02

Copy Agent

Writes every text node: headlines, body paragraphs, CTA labels, list items, meta descriptions. Tone-aware and industry-specific. The Copy Agent receives the layout blueprint and fills in content that fits the structure.

03

Style Agent

Applies Stride CSS classes to every element: colors, spacing, typography, responsive behavior, backgrounds. The Style Agent sees both the layout and copy, ensuring visual consistency across the entire page.

Layout Agent

Structural intelligence.

The Layout Agent is trained on thousands of high-converting page structures across industries. It understands that a dental clinic landing page needs a hero, services grid, testimonials, and contact form — in that order.

  • Section planning: Determines how many sections the page needs and what purpose each serves.
  • Grid decisions: Chooses column counts, widths, and nesting depth based on content density.
  • Element selection: Picks element types (heading, text, button, image, list, form) for each content slot.
  • Schema output: Emits a JSON tree following the Gallop element schema exactly.
// Layout Agent output (excerpt)
{
  "sections": [
    {
      "purpose": "hero",
      "rows": [{
        "columns": [
          { "width": "1/2",
            "elements": [
              "heading", "text",
              "button_group"
          ]},
          { "width": "1/2",
            "elements": ["image"]
          }
        ]
      }]
    },
    {
      "purpose": "services",
      "rows": [{
        "columns": 3,
        "elements": ["card", "card",
          "card"]
      }]
    }
  ]
}

Copy Agent

Content that converts.

The Copy Agent doesn't just fill in placeholder text. It writes for the specific industry, audience, and tone defined in the prompt. Headlines follow conversion copywriting patterns. CTAs use proven action language.

  • Industry awareness: Knows a dental clinic CTA should say "Book Your Visit" not "Get Started."
  • Tone control: Professional, casual, technical, or playful — driven by the prompt or settings.
  • SEO meta: Generates title tags and meta descriptions optimized for search.
  • Content length: Calibrates word count per element type (short headlines, detailed paragraphs).
// Copy Agent output (excerpt)
{
  "hero.heading": "Your Smile
    Deserves Expert Care",
  "hero.subtext": "Bright Smile
    Dental has served Austin
    families for over 15 years.
    Book your visit today.",
  "hero.cta_primary": "Book Now",
  "hero.cta_secondary": "Our
    Services",
  "services.card_1.title":
    "General Dentistry",
  "services.card_1.body":
    "Cleanings, exams, and
    preventive care for the
    whole family.",
  "meta.title": "Bright Smile
    Dental | Austin TX",
  "meta.description": "Trusted
    dental care in Austin..."
}

Validation

Schema-first generation.

Every agent's output is validated against the Gallop element schema before the outputs are merged. Invalid structures are rejected and regenerated — the user never sees a broken page.

InputNatural language prompt describing the desired page.
PipelineThree specialized agents run in parallel, each constrained by JSON schema.
ValidationOutput is validated against the element schema. Type mismatches, invalid nesting, and missing required fields are caught.
RetryFailed validations trigger automatic regeneration of the failing agent's output. Up to 3 retries per agent.
MergeValidated outputs from all three agents are merged into a single page schema: layout + content + styling.
OutputStandard WordPress content JSON. Native blocks, no shortcodes, no proprietary format. Ready for the block editor.

Performance

Speed through parallelism.

42s average generation

A typical 6-section landing page generates in 42 seconds. Complex pages with 10+ sections average 65 seconds. Parallel execution is the key — agents don't wait for each other.

3x agent parallelism

Layout, Copy, and Style agents run simultaneously. Total generation time is determined by the slowest agent, not the sum of all three. This cuts wall-clock time by roughly 60%.

98% first-pass validation

Schema constraints dramatically reduce hallucination. 98% of generations pass validation on the first attempt. Only 2% require a retry, and retries are fast because only the failing agent re-runs.

100% valid output

Every page that ships to your WordPress site has passed full schema validation. No broken HTML, no invalid nesting, no orphaned elements. Production-ready on delivery.

Early Access

See the engine in action.

Get early access when we ship.

Get Early Access

Join the waitlist and be the first to try Gallop Builder.

No spam. Unsubscribe whenever.