You can get a New York City building's DOB permits from the NYC Open Data building permits API with two HTTP calls and no key: turn the address into a BIN with the city's GeoSearch service, then query two Socrata datasets by that BIN, one for the old BIS system and one for DOB NOW. Sort stably when you page, keep dedupe inside each dataset, and treat placeholder BINs as a warning. This post gives the script, a field map and the traps we recorded on November 20, 2026.
Why two datasets
NYC Open Data publishes DOB permits as separate tables, so one query cannot return a building's whole history. ipu4-2q9a (DOB Permit Issuance) says each row is the life cycle of one permit for one work type. rbx6-tga4 (DOB NOW: Build – Approved Permits) lists approved DOB NOW permits except Electrical, Elevator and Limited Alteration Application permits, which have their own datasets. Different keys, different vocabularies, different date types.
Step 1: address to BIN with GeoSearch
GeoSearch is a free city service. Its documentation says /v2/search takes a text parameter and returns 10 places by default, with size to change that. Every returned feature can carry a BIN, so one address can give several.
import requests GEO = "https://geosearch.planninglabs.nyc/v2/search" def address_to_bins(text): r = requests.get(GEO, params={"text": text, "size": 5}, timeout=30) r.raise_for_status() bins = {} for f in r.json()["features"]: p = f["properties"] b = p.get("addendum", {}).get("pad", {}).get("bin") if b: bins.setdefault(str(b), p["label"]) # one entry per BIN return bins Dedupe by BIN here. Our first run did not, and the same BIN was queried three times.
Step 2: query both datasets, with paging
Socrata returns rows in no promised order. Its ORDER BY page says that if you page you should provide $order, or at least $order=:id, to keep the order stable. Paging uses $limit and $offset.
SODA = "https://data.cityofnewyork.us/resource/{}.json" SPECS = { "ipu4-2q9a": {"bin": "bin__", "date": "issuance_date", "key": ["job__", "job_doc___", "work_type", "permit_sequence__"], "fields": ["job__", "job_doc___", "work_type", "permit_sequence__", "issuance_date", "permit_status"]}, "rbx6-tga4": {"bin": "bin", "date": "issued_date", "key": ["work_permit"], "fields": ["work_permit", "work_type", "issued_date", "permit_status"]}, } def fetch(dataset, bin_, page=50000): s, rows, offset = SPECS[dataset], [], 0 while True: params = {"$select": ",".join(s["fields"]), "$where": f"{s['bin']}='{bin_}'", "$order": ":id", "$limit": page, "$offset": offset} r = requests.get(SODA.format(dataset), params=params, timeout=90) r.raise_for_status() batch = r.json() rows += batch if len(batch) < page: return rows offset += page Note the column names: the BIS table spells the building column bin__ with trailing underscores, DOB NOW spells it bin. Both are text.
Step 3: merge, deduping inside each dataset only
The two datasets share no permit number, so a cross-dataset dedupe has nothing to match on. Dedupe each one on its own key, then tag rows with their source and sort by date.
from datetime import date, datetime def dedupe(dataset, rows): seen, out = set(), [] for x in rows: k = tuple(x.get(f, "") for f in SPECS[dataset]["key"]) if k not in seen: seen.add(k) out.append(x) return out def to_date(value): for fmt in ("%m/%d/%Y", "%Y-%m-%dT%H:%M:%S.%f"): try: return datetime.strptime(value or "", fmt).date() except ValueError: pass return None # empty or unparseable: keep the row, flag it def lookup(address): merged = [] for bin_, label in address_to_bins(address).items(): for ds, s in SPECS.items(): for x in dedupe(ds, fetch(ds, bin_)): merged.append({"bin": bin_, "dataset": ds, "work_type": x.get("work_type", ""), "status": x.get("permit_status", ""), "date": to_date(x.get(s["date"]))}) merged.sort(key=lambda m: (m["date"] is not None, m["date"] or date.min), reverse=True) return merged The BIS key includes work_type because that table keeps one row per work type per permit sequence, and some BIS rows have an empty work_type. Leave the empty string in the key rather than dropping those rows.
Field mapping cheat sheet
From the two datasets' column metadata, pulled November 20, 2026:
| Meaning | BIS ipu4-2q9a | DOB NOW rbx6-tga4 |
|---|---|---|
| Building | bin__ (text) | bin (text) |
| House number | house__ | house_no |
| Street | street_name | street_name |
| Job | job__ | job_filing_number |
| Permit sequence | permit_sequence__ | sequence_number |
| Permit number | none; use the composite key above | work_permit |
| Work type | work_type (codes such as MH, PL) | work_type (names such as Mechanical Systems) |
| Status | permit_status | permit_status |
| Issue date | issuance_date (text) | issued_date (calendar date) |
| Tax lot | bbl (number) | bbl (number) |
| Community board | community_board (text) | community_board (number) |
Work-type labels line up for only a handful: MH with Mechanical Systems, PL with Plumbing, BL with Boiler Equipment, FS with Sprinklers, SD with Sidewalk Shed, CC with Curb Cut. We left OT, EQ, FB, FA, FP, NB and SP unmapped, and DOB NOW categories such as Construction Fence and Supported Scaffold have no BIS counterpart in the column text. Map what the metadata supports and write "unmapped" for the rest.
Data quality gotchas (as of November 20, 2026)
Counts are whole-table numbers from that day and will drift.
- Table sizes. BIS held 3,991,416 rows and DOB NOW 1,032,205. DOB NOW has 618,323 distinct
work_permitvalues, so that field repeats across rows and is not a row id. - BIS dates are text.
issuance_dateis a text column, so a server-side year extract failed with HTTP 400 andmax()would sort strings. 20,950 BIS rows have no issuance date at all. Parse locally. - Placeholder BINs. 38 DOB NOW rows carry one of 1000000, 2000000, 3000000, 4000000 or 5000000. The BIS table had none. No row in either table had a null BIN.
- Paging without a stable sort lies. Our first per-year totals came from grouped, paged queries. For BIS they summed to 4,866,751 against a table count of 3,991,416, so we discarded them. We did not get a reconciled per-year table, so this post has none.
- Freshness. The dataset metadata gave BIS a last row update of 2026-11-18 00:38 UTC and DOB NOW 2026-11-18 19:44 UTC.
Worked example: one Brooklyn address
We ran the collection script on 1021 Bedford Avenue and saved sample-output.csv. GeoSearch labelled the top match "1021 BEDFORD AVENUE, Brooklyn, NY, USA", BBL 3017887506, BIN 3414286.
| Dataset | Rows | First issue date | Last issue date | Mechanical |
|---|---|---|---|---|
BIS ipu4-2q9a | 11 | 2019-06-18 | 2021-11-02 | 0 (no MH) |
DOB NOW rbx6-tga4 | 5 | 2019-01-29 | 2021-10-01 | 0 (no Mechanical Systems) |
The BIS rows were OT ×4, PL ×2 and five with an empty work type, all ISSUED. The DOB NOW rows were Construction Fence ×2, Supported Scaffold ×2 and Sidewalk Shed ×1, all Signed-off. Mechanical was zero on both sides, so these two datasets list no mechanical permit for this BIN.
The same CSV shows the trap. That script queried every GeoSearch match, not just the first: the file holds 87 rows, only 43 of them distinct. Three nearby addresses (1362, 1366, 1368 Bedford) share BIN 3030276 and repeated the same 13 permits. Three others (1334, 1367, 1391) came back with BIN 3000000, a placeholder, and each returned the same 9 DOB NOW rows. Those rows belong to no single building. The script in this post dedupes by BIN, and you should also drop placeholders before reading anything into the result.
Free open data vs paid permit APIs
Paid permit APIs exist, and we did not open any vendor page for this post, so we name none and quote no price or freshness. The trade is general. Open data costs nothing and you see the raw columns, but you own the BIN matching, paging, date parsing and cleanup described above, and you track the dataset's refresh yourself. A paid service may do some of that for you; read its own documentation for what it covers and how current it is.
On limits: the application token page says requests without a token share a pool by IP address and may be throttled with a 429. With a token, usage is attributed to your application and gets its own pool, sent in the X-App-Token header. The SoQL overview adds that SODA3 requests must be authenticated or carry a valid token. We made six unauthenticated requests to the /resource/ endpoints on November 20 and none returned a 429; that is six requests, not a measured limit. For anything scheduled, register a token.
Edge cases a permit lookup has to handle
Any address-based permit lookup meets these three. Each is paraphrased from the HVACLookout tool's own page, read on November 20, 2026:
- Several matches. The page says that when an address matches several lots or BINs, up to 10 candidates appear with address, BBL and BIN and the user picks one.
- Condo lots. It says condo buildings have a billing BBL and separate unit lots, and a permit may cite either, so try the other address or BBL.
- Stale "Permit Issued". It says DOB's data often keeps that status after the work ends without a recorded sign-off, so read the status next to the issue date.
If you would rather not write code
The NYC permit lookup tool takes an address or a 10-digit BBL and shows a building's permit timeline. It reads one dataset only, the DOB NOW approved permits (rbx6-tga4); it does not read the BIS dataset, so older BIS-era permits are not on it.
FAQ
is there a free API for NYC building permits
Yes. The two datasets above sit behind Socrata's SODA API on data.cityofnewyork.us, and a JSON request needs no payment. Use $where on bin or bin__, and add an app token once the job is scheduled.
how to get DOB permits by BIN
Query each dataset with $where=bin__='<BIN>' (BIS) and $where=bin='<BIN>' (DOB NOW), add $order=:id, then merge on your side. A BIN is seven digits; reject the five placeholder values.
difference between DOB permit issuance and DOB NOW dataset
In the data: one has composite keys and text dates, the other a single work_permit field and real date types. One row is a work-type permit life cycle; the other lists approved DOB NOW permits except three types with their own datasets.
how do I get NYC building permit data by address with an API
GeoSearch first for the BIN, then the two SODA calls. Filtering on house__/house_no plus street name also works, but you must match each table's street-name text yourself, so BIN is the simpler join.
NYC Open Data DOB permits API do I need an app token
Not for a first query: our six unauthenticated calls worked. Socrata's docs describe token-less traffic as a shared, throttleable pool, so register one for anything repeated.
why do I get duplicate or missing permits when querying NYC DOB open data by BIN
Duplicates: several addresses on one BIN, or unstable paging. Missing: placeholder BINs, or the history sitting in the other dataset. Order by :id, dedupe per dataset, query both.
Sources
- NYC Open Data, DOB Permit Issuance (ipu4-2q9a)
- NYC Open Data, DOB NOW: Build – Approved Permits (rbx6-tga4)
- Socrata, Application Tokens
- Socrata, The ORDER BY Clause
- Socrata, The OFFSET Clause
- Socrata, SoQL Queries
- NYC Planning Labs, GeoSearch documentation
- Data and script notes (pulled November 20, 2026): available on request at contact@hvaclookout.com


