NYC DOB Permits — Cleaned & Tagged Building Permit Data

NYC Department of Buildings permit records, cleaned, deduplicated and rule-tagged for AI agents and analysts. Query live, or pull a tagged historical snapshot.

Most building-permit scrapers hand you a lead: an address and a date. This one gives you the analysis fields — estimated_cost and the full description are populated, plus five tag dimensions you can filter on.

---

What you get

One dataset item per permit:

Field	Description
permit_id	Job filing number (stable identifier)
status	Normalised: issued, reissued, in_progress, revoked, signed_off
application_type	Source work type (e.g. Sidewalk Shed, Solar, Plumbing)
work_type	Tag — canonical work type (scaffolding, solar, fire_protection, …)
building_type	Tag — residential, commercial, industrial, mixed, institutional
use_type	Tag — housing, retail, office, education, health, …
value_category	Tag — low (<$100k), mid ($100k–$1M), high (≥$1M)
estimated_cost	Estimated job cost in USD (integer)
description	Full work description text
date_lodged / date_decided	Approval / issuance dates (ISO-8601)
borough, postcode, property_address	Location
bin, bbl, latitude, longitude, nta, community_board, council_district	Identifiers and geography
tags	Flattened list of all tags, for easy filtering
tag_confidence	Per-tag confidence: 1.0 deterministic, 0.7 keyword-derived
license_note	Provenance, on every row

36 fields in total.

Why this one

• estimated_cost populated. The cheap permit scrapers on this Store return it empty.
• Tagged, not raw. Five dimensions pre-computed, so an agent can filter without a second pass.
• No personal data. The NYC source carries applicant, owner and filing-representative names and licence numbers. Those columns are dropped at harvest time — this Actor never emits them. Every row carries an attribution + no-raw-resale licence note.
• Deterministic tags. Rule-based, versioned (tag_version: 3.0-rule), reproducible — not an LLM guess.

Modes

Mode	What it does
live (default)	Queries NYC Open Data in real time. Freshest data; filter by date, work type, borough and cost.
snapshot	Serves the bundled clean/tagged historical sample (2,000 rows). No network, instant, deterministic.

Input

{
  "mode": "live",
  "city": "nyc",
  "dateFrom": "2026-01-01",
  "dateTo": "",
  "workTypes": ["Solar", "Roofing"],
  "borough": "BROOKLYN",
  "minEstimatedCost": 100000,
  "maxItems": 1000
}

• workTypes must match the source values (e.g. Solar, Plumbing, General Construction, Sidewalk Shed, Structural, Foundation).
• maxItems is a hard cost cap — you are charged per record returned.

Output example

{
  "permit_id": "B00002006-I1",
  "status": "signed_off",
  "application_type": "Sidewalk Shed",
  "work_type": "scaffolding",
  "value_category": "mid",
  "estimated_cost": 103000,
  "description": "INSTALLATION OF SIDEWALK SHED AS PER PLANS...",
  "borough": "Brooklyn",
  "postcode": "11207",
  "property_address": "270 WORTMAN AVENUE",
  "date_lodged": "2017-11-08",
  "date_decided": "2018-09-13",
  "tags": ["scaffolding", "value_mid"],
  "tag_confidence": {"work_type": 1.0, "value_category": 1.0},
  "license_note": "Public record from NYC DOB (data.cityofnewyork.us)..."
}

Pricing

Pay per event. You are charged per permit record returned. Set maxItems to cap your spend before you run.

Data source and licence

Source: NYC DOB NOW: Build – Approved Permits (rbx6-tga4), published under the NYC Open Data Law, which permits commercial reuse. Attribution required. Our cleaning, normalisation and tagging are our own work product; redistribution of raw source rows is not permitted.

No personal data. PII columns are excluded at harvest time.

Notes and limits

• v1 covers New York City only. More cities are on the roadmap.
• building_type and use_type are keyword-derived from the free-text description, so they are populated where the text supports it and null otherwise (confidence 0.7). work_type and value_category are deterministic.
• The live mode reads a public Socrata endpoint; very large maxItems values will take longer.