# How the API works

What happens to a request, and why the API can swap its engines without breaking your client.

## One contract over interchangeable engines

Routing and geocoding are done by separate engines. `/v2` puts one contract in front of them, so your client talks to the same request and response shapes whichever engines are running. `GET /v2/info` tells you which ones are.

## What happens to a request

1. **The gateway checks your key and your rate limit.** A missing or invalid key stops here with `401`, and too many requests stop here with `429`.
2. **The API validates the request** against the published schemas. Anything invalid stops with a `400`.
3. **An engine does the work.** The API translates your request into that engine's format.
4. **The API shapes the reply** into the contract, so units, field names and errors are the same every time.

## What this means for your code

- **Write to the contract.** Use the documented fields, not engine-specific behaviour. See [API rules](../reference/api-rules.md).
- **Expect some answers to differ between engines.** The contract promises the shape, not identical routes or places.
- **Check what's offered.** A capability appears in `GET /v2/info` only if its endpoint works here.

## Data

Roads and places come from OpenStreetMap, as a snapshot rather than a live feed. See [Coverage and data](coverage.md).