What's in the data, how it's organized, and how to plug it into your stack.

The data

Four layers, one project graph: what's required, who runs it, what form to file, and where the jurisdiction lines fall.

How it's organized

Everything hangs off a single primitive: the permit definition. Each one carries the logic that decides whether it applies to a given project, plus the metadata a filer needs.

// A permit definition — the core object
{
  id:      'cgp',
  name:    'NPDES Construction General Permit (CGP / SWPPP)',
  agency:  i => ST_ENV[i.state] || 'State Environmental Agency',
  level:   'state',        // federal | state | local
  time:    '7–14 days',     // typical review window
  cost:    '$0 filing; SWPPP required',
  chk:     i => (i.over1ac || i.ms4_area) ? 'req' : null,  // the trigger rule
  why:     () => 'Ground disturbance >1 acre triggers the CGP.',
  trees:   ['consultant', 'owner']   // who typically holds it
}

The four data layers

• Permit catalog — the 241 definitions above. A federal + base set applies nationwide; per-state extensions (currently AZ, CA, CO, FL, GA, IL, NC, NY, TX, WA) layer in state-specific programs.
• Forms directory — maps each permit id to its official application form(s), tagged pdf (stable direct download) or page (agency page when the PDF URL is volatile). Re-verified monthly.
• Agency & jurisdiction data — contact records keyed to permit IDs, plus a ZIP→city→AHJ resolver so "local grading permit" becomes "City of Boise Public Works."
• State environmental profiles — the narrative + structured regime for each state: construction stormwater, §404 assumption status, wetlands, water rights, land-use review (e.g. Act 250, CEQA), and species overlays.

How it's queried

A project isn't a search box — it's a set of facts. EntitleHQ evaluates every permit's chk() rule against those facts and returns a ranked, deduplicated permit set.

// Query = facts in, permit set out
input = { state:'ID', projectType:'land_dev', over1ac:true, waterway:true, floodplain:true }

result = PERMITS
  .map(p => ({ ...p, status: p.chk(input) }))   // evaluate every rule
  .filter(p => p.status)                          // keep req + evl
  .map(p => resolveAgencyFormAHJ(p, input))       // attach agency, form, office
  .sort(byLevelThenTimeline)

Because triggers are declarative functions rather than a static list, the same project entered in Idaho and Florida returns different sets — Idaho surfaces the state CGP + ACHD ROW; Florida adds the SWFWMD/SJRWMD Environmental Resource Permit.

The AI layer

On top of the structured model sits an AI layer that reads documents, drafts applications, and executes in-portal. It always operates against the project's live permit set — never a blank prompt.

• Scope & explain (chat) — plain-English Q&A grounded in the active project's permits and agencies.
• Document read (permit_doc) — upload a plan set, agency letter, or issued permit; the model extracts status, dates, fees, conditions, and updates the tracker.
• Pre-fill (prefill) — maps project data onto an application's fields and flags what's still missing.
• Fill PDF (fill_pdf) — writes the mapped values into the actual government PDF via the form's field schema.
• Agentic execution (agent) — tool-use loop that can make changes in-portal; every action is shown and reversible, so you stay in QA.

AEC integrations

EntitleHQ is designed to sit inside the stack you already run, not replace it. Below is the integration surface — what's live, what's in beta, and what we'll build with design partners. Live Beta Planned

Accuracy & sourcing

The data is only worth what it's right about. Three commitments hold the model honest.

• Authoritative sources. Every form links to the issuing agency's official page. State profiles cite the governing statute or program (e.g. NPDES CGP number, Act 250, CEQA).
• Human-in-the-loop. AI-drafted and AI-filed content is always shown and reversible before it leaves the portal — the app is your QA surface, not an autopilot you can't see.
• Built by a licensed civil P.E. The trigger logic reflects how permitting actually works on real land-development projects, not a scraped keyword index.