# AI Kit Data

> Offline NHTSA vPIC VIN decoding at the edge, plus related US vehicle data. No API key. Independent project by AI Kit LLC, not affiliated with any U.S. government agency.

Decode a VIN with one GET request, for example:

    curl -s https://data.ai-kit.net/api/v1/vin/1HGCM82633A004352

The response has `data.make`, `data.model`, `data.modelYear`, `data.isCleanDecode` and `data.attributes` (every decoded vPIC variable). Check `data.errorCodes` before trusting a result: partial or invalid VINs are decoded as far as possible and flagged. Requests are limited to 120 per minute per client; a 429 carries Retry-After.

## API

- [OpenAPI 3.1 description](https://data.ai-kit.net/openapi.json): every endpoint, parameter and response
- [API reference](https://data.ai-kit.net/docs): the same, as a web page
- [VIN decoder](https://data.ai-kit.net/vpic): interactive decoder, single and batch

## Data and accuracy

- [Data pipeline and verification](https://github.com/AIKitLLC/nhtsa-edge-api/blob/main/docs/DATA.md): the decoder is compared with NHTSA's own SQL functions on the original dump, and with the live vPIC API
- [Known defects in NHTSA's dump](https://github.com/AIKitLLC/nhtsa-edge-api/blob/main/docs/NHTSA-ERRATA.md)
- [Source code, MIT](https://github.com/AIKitLLC/nhtsa-edge-api)

## Limits

- Only vPIC decoding is fully offline. Recalls are proxied to NHTSA and cached; EPA and RDW enrichment is at model level, not per vehicle.
- Results can differ from the live vPIC API where NHTSA has published newer data than the monthly dump the service serves; every response states its data version in `X-Vpic-Data-Version`.