Page Intelligence · SiteLens Cloud/API
Structured page understanding for agents—not a DOM dump.
Page Intelligence renders a public URL in Chromium and returns a PageModelV1 with ranked actions and inferred workflow stages. Agents use it to decide what to click, what matters in the hero, and how a product explains itself—then prove outcomes with AI Agent Verification flows.
What Page Intelligence is
Shipped
Page Intelligence sits on top of SiteLens page extraction: headings, actions, regions, navigation landmarks, and text blocks are collected from a real render, then heuristics rank the actions that matter for conversion and navigation. The result is agent-friendly JSON—primaryActions for hero CTAs, secondaryActions for supporting links, navigationActions for chrome, and workflowStages for “how it works” style journeys.
Plan before you automate
Agents read confidence-scored summaries instead of guessing selectors from raw HTML—useful when generating SiteLens flows or scoping QA.
Same engine as QA
Chromium render semantics align with SiteLens QA—what you analyze is what you can later verify with local or Cloud/API runs.
Intelligence fields
Each field is an array on PageModelV1 with confidence scores and stable ids.
primaryActions
Hero-weighted CTAs—“Get started”, “Sign up”, “Try free”, and similar conversion labels ranked above nav chrome. Scoring considers button vs link, viewport position, hero region overlap, and label patterns.
secondaryActions
Supporting actions—“Learn more”, docs links, pricing, contact—deduplicated against primaries so agents see a clear primary/secondary split.
navigationActions
Header/nav and landmark links—sign in, docs, product sections—merged from action context and navigation landmarks for wayfinding plans.
workflowStages
Inferred steps from numbered lists, “How it works” sections, clustered headings, and action verbs—useful when drafting multi-step verification flows.
Illustrative response fragment
PageModelV1{
"success": true,
"version": 1,
"page": { "title": "Example SaaS", "url": "https://example.com/" },
"primaryActions": [
{ "id": "pa-1", "label": "Start building", "kind": "button", "confidence": 0.91, "source": "primary-heuristic" }
],
"secondaryActions": [
{ "id": "sa-1", "label": "View documentation", "kind": "link", "confidence": 0.74, "source": "secondary-heuristic" }
],
"navigationActions": [
{ "id": "na-1", "label": "Pricing", "kind": "link", "confidence": 0.68, "source": "navigation-action" }
],
"workflowStages": [
{ "id": "ws-1", "label": "Connect your data", "confidence": 0.72, "source": "numbered-text" },
{ "id": "ws-2", "label": "Customize dashboards", "confidence": 0.70, "source": "numbered-text" }
],
"actions": [ "... full action catalog with selector hints ..." ],
"regions": [ "... hero, bands, landmarks ..." ]
}
API overview
Hosted POST /analyze/page returns PageModelV1 for a public HTTPS URL. Requires a workspace API key (Authorization: Bearer sitelens_sk_…). Optional request flags control screenshot, regions, and text block inclusion.
curl -s -X POST https://api.sitelensapi.com/analyze/page \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sitelens_sk_<your_workspace_key>" \
-d '{"url":"https://example.com","options":{"includeRegions":true,"includeTextBlocks":true}}' | jq '.primaryActions, .workflowStages'
Relationship to Theme API
ComplementarySiteLens Theme answers “how does this page look?”—colors, typography, and confidence-scored design tokens via POST /analyze/theme. Page Intelligence answers “what can a user do here?”—CTAs, nav, and workflow structure via POST /analyze/page. Together they support match-my-site embeds (Theme) and agent planning + verification (Page Intelligence + flows).
- Theme → visual tokens for adapters and white-label UX
- Page Intelligence → action and workflow summaries for automation
- Both use the same workspace key and hosted Chromium runtime
Relationship to AI Agent Verification
Plan → prove
Page Intelligence is the planning layer; AI Agent Verification is the proof layer. An agent can call /analyze/page to discover primary CTAs and workflow stages, draft a .sitelens/flows JSON, then run sitelens_run_local_flow from Cursor to verify the journey in a real browser. The monorepo includes page-intelligence-marketing-smoke—a flow that exercises marketing homepage CTAs after Page Intelligence-informed planning.
Related
Contact & onboarding
Page Intelligence access, agent workflows, or API questions — same short form everywhere.