NHTSA VIN Decoder & Vehicle Recalls
Pricing
from $0.03 / 1,000 nhtsa vehicle records
NHTSA VIN Decoder & Vehicle Recalls
US NHTSA vehicle data (public open data, no key): decode a VIN into 69 structured fields - make, model, year, body, engine, plant, safety equipment - and pull the safety recalls for a vehicle or by campaign number, with the remedy and the recall date.
Pricing
from $0.03 / 1,000 nhtsa vehicle records
Rating
0.0
(0)
Developer
Wenhao Yang
Maintained by CommunityActor stats
0
Bookmarked
2
Total users
1
Monthly active users
8 hours ago
Last modified
Categories
Share
NHTSA Vehicles - VIN Decode & Safety Recalls
Decode any VIN into its full factory specification, and pull NHTSA safety recalls for a vehicle or a campaign - from two free US government services, in one actor.
Low cost
From $0.00005 per record, down to $0.00003 at Gold Pay-per-event: you are charged per record delivered, and nothing for the query.
What you get
| Corpus | One record is | Depth |
|---|---|---|
vin | one decoded VIN | 69 flattened vehicle fields, plus the complete 140-variable decode |
recalls | one recall campaign for one vehicle | campaign number, component, summary, consequence, remedy, units affected, date |
The two are different grains on purpose. A VIN decode is one row per VIN. A recall is one row per campaign - a vehicle has many, and a campaign that covers several models comes back as several rows sharing one campaign number.
VIN decode
NHTSA's decoder answers with 140 Variable/Value pairs for a single VIN. That is more than a flat row can hold usefully, so 69 are flattened onto the record - identity, plant, dimensions and weight, engine and electrification, transmission, seating and the safety-equipment block - and the complete 140-pair answer is kept verbatim in allVariablesJson. Nothing is lost to the curation.
A real decode, 1HGCM82633A004352:
Values arrive as strings even when numeric (2.998832712, 240, 5), and a variable NHTSA holds no data for is '' - which is not the same as "the vehicle lacks it", and is not claimed to be.
Safety recalls
- By vehicle - give
make,modelandmodelYear(all three; the source rejects any subset).honda/accord/2003returns 24 recalls. - By campaign - give a campaign number.
19V182and19V182000both resolve to campaign19V182000, which covers 93 vehicle rows (Acura CL, MDX, TL and more), each carrying the same campaign number and its own model.
Decode -> recalls in one run. Set includeRecalls=true with corpus=vin and each VIN's decoded make/model/year is used to fetch its recalls, emitted next to the decode record. One caveat, stated on every record as recallScope: "make-model-year": NHTSA publishes recalls by make/model/year, not by VIN. Some campaigns cover a model-year range, so a VIN's recall set is its make/model/year's recalls - a superset of that exact VIN's campaigns. It is honest about what it is, not silently presented as VIN-exact.
Modes
rows(default) - one record per VIN, or per recall campaign.aggregate-corpus=recallsonly: one count row per group, bycomponent,manufacturer,modelYearorrecallMake. Counted over the recall set NHTSA returned for the query (a make/model/year's recall set is small - the largest measured was 24 rows).
Example inputs
Decode several VINs in one run - corpus=vin; each VIN is one request, and maxVins caps the batch.
One VIN plus its recalls - includeRecalls=true emits the recalls next to the decode record.
One recall campaign - campaignNumber takes precedence over make/model/modelYear.
A 2003 Accord's recall mix - mode=aggregate counts the fetched recall set by component.
Example output
One decoded VIN - corpus=vin, vins=["1HGCM82633A004352"]; the 140-variable decode is carried verbatim in allVariablesJson:
corpus=recalls, mode=aggregate, groupBy=component for honda / accord / 2003 - one count row per component:
What gets recalled on a 2003 Accord - corpus=recalls, make=honda, model=accord, modelYear=2003, mode=aggregate, groupBy=component:
Notes on the source
ReportReceivedDateisDD/MM/YYYY, notMM/DD/YYYY. Measured across two endpoints: 35 values with the first component greater than 12, and none with the second greater than 12 -27/06/2019,30/03/2017. Read as US dates it would mis-date every recall, and06/03/2019is unreadable without the rule. It is normalised to ISOYYYY-MM-DDon output.Error Codeis a comma-joined set, not one number - a clean VIN is0, a malformed one is6,7,11,400, spelled out indecodeErrorText. An invalid VIN is not an error from the API: NHTSA answers HTTP 200 with an error code, so a bad VIN is a decoded record that says so.- A 400 from the recalls feed means "not in NHTSA's vehicle table", not "this car has no recalls". Those are different statements, so a bad combination is reported back naming the make/model/year rather than served as silence. A vehicle with genuinely no recalls returns 200 with
Count: 0. - Manufacturer casing is inconsistent in the source - the same company appears as both
Honda (American Honda Motor Co.)andHONDA (AMERICAN HONDA MOTOR CO.)across records. Group by it and both appear; that is the source, not a normalisation this actor applies. - Recall counts are verified complete. The feed has no pagination, and every response is checked so that
Countequals the rows returned - if NHTSA ever starts truncating, the run says so instead of serving a short recall history as a whole one.
Source
NHTSA (US National Highway Traffic Safety Administration), two public services: the vPIC VIN decoder at vpic.nhtsa.dot.gov and the recalls API at api.nhtsa.gov. Public US government data, no login, no key.
Both are lookup services, not bulk files: one VIN decode is one request, and there is no endpoint that enumerates vehicles. So the request count is bounded by design - a hard ceiling of 500 VINs per run, an 8-wide worker pool, pacing jitter, and retry with backoff on 429/5xx. Measured from the build box on 2026-09-13: no throttling was observed at 50 concurrent decodes (13 req/s), and none in 100 sequential calls - the ceiling reached was our own concurrency, not NHTSA's. That is not read as "no limit exists": the bounds stay in place because an actor where request count equals user input length is one that can be aimed at the API with a 100,000-VIN list.
A VIN that cannot be fetched after retries is not dropped and not emitted blank - it comes back as a record with decodeErrorCode: FETCH_FAILED and the reason in error, so a short answer is visible as short. If failures look systemic (half of a batch), the run fails instead, because that is the source being down and not a bad VIN.
Output
Every record carries the same key set - all 98 keys - whichever corpus or mode produced it, with '' for absent values, so a dataset column never shifts row to row. The envelope (platform, source, corpus, recordKind, mode, truncated, error, vin, recallScope, groupKey, groupCount, groupBy) is identical across corpora; groupKey/groupCount/groupBy are '' outside aggregate mode.
Related actors
- FMCSA Carrier Safety - the federal motor-carrier register: inspections, violations, operating authority and out-of-service orders for the carriers these vehicles run under.
![VIN Decoder [💰$2/1K] | Vehicle Specs | Safety Recalls | NHTSA avatar](https://images.apifyusercontent.com/yCkssfr7yORAvMfRa8b43g7sez24hEZ-SgXY-c4SUtc/rs:fill:76:76/cb:1/aHR0cHM6Ly9hcGlmeS1pbWFnZS11cGxvYWRzLXByb2QuczMudXMtZWFzdC0xLmFtYXpvbmF3cy5jb20vZHRSN3FuVGJRck5pU0NtajctYWN0b3ItN2JoZDc1TkdqUG1CTlBjdG0tb3REbWZSOVBSYy1ibG9nLXZpbi1udW1iZXItbG9va3VwLWhlcm8ucG5n.webp){kind=avatar}
