Yappy API sandbox
Everything under /api/v1/sandbox behaves exactly like the live API and touches nothing real. Use it to build and test an integration, and to prove your error handling works before you find out in production.
Base URL
https://yappy.biz/api/v1/sandboxEvery path under /api/v1 exists under it. GET /api/v1/sandbox/releases/latest mirrors GET /api/v1/releases/latest, with fixture data instead of live data:
curl -s https://yappy.biz/api/v1/sandbox/releases/latest{
"object": "release",
"version": "0.0.0-sandbox",
"notes": "Fixture release. Deterministic, never changes, safe to assert against.",
"published_at": "2026-01-01",
"download_url": "https://yappy.biz/api/v1/sandbox/download",
"sandbox": true
}Every sandbox response carries "sandbox": true in the body and an X-Yappy-Environment: sandbox header, so a misconfigured client that is pointing at the wrong base URL is obvious in one glance at the response rather than three hours into an incident.
Deterministic fixtures
Sandbox data never changes. That is the point: you can assert on exact values in a test suite without it breaking the day a real release ships. Version is always 0.0.0-sandbox, the docs corpus is a fixed three pages, and the FAQ is a fixed two entries.
Forcing errors
Every error path is reachable on demand with the ?simulate= parameter, so you can test your retry and backoff logic without waiting for a real failure:
simulate | Response |
|---|---|
rate_limited | 429 with Retry-After: 3 and the rate-limit headers |
invalid_request | 400 with a populated errors[] array |
not_found | 404 |
invalid_token | 401 with a WWW-Authenticate challenge |
insufficient_scope | 403 naming the missing scope |
internal_error | 500 |
upstream_unavailable | 503 with Retry-After |
slow | A successful response after a two-second delay |
curl -si "https://yappy.biz/api/v1/sandbox/releases?simulate=rate_limited"Async jobs
POST /api/v1/sandbox/exports returns a 202 and a job that reports queued, then running, then completed across successive polls — so you can exercise a polling loop that actually loops, instead of one that gets completed on the first try and never tests the interesting path.
Rate limits
The sandbox has its own bucket and does not consume your production allowance. It is limited more tightly on purpose — 60 requests per minute — because a test suite that needs more than that is testing our infrastructure rather than your integration.
What is not simulated
There is no sandbox for dictation, because there is no production endpoint for dictation either — speech recognition runs on the user's Mac. The sandbox mirrors the metadata API and nothing else.