FAQ
The details, without the ambiguity
Straight answers on data handling, determinism, performance, and integration.
Do you store the titles I send?+
No. Titles are processed in memory only and are never persisted. The API is stateless. Nothing about your request is retained after the response is returned.
Are results deterministic?+
Yes. Classification is pattern-based, not model-based. The same title with the same pattern version always produces exactly the same result, which makes results safe to cache and audit.
How is the API versioned?+
The URL path (/v1/) versions the request and response shapes; breaking changes ship under a new path version. Separately, every response includes taxonomy and patterns version strings. A patterns bump means the same title may classify differently than before, so rerun cached classifications if you depend on consistency.
What are the rate limits?+
A free trial key allows 100 single-title requests per day and cannot use the batch endpoint. A Standard key allows 10 requests per second with a burst of 20, and 100,000 requests per day. A batch of up to 500 titles counts as one request.
What happens when I exceed the limits?+
You receive a 429 Too Many Requests response. Back off and retry. Limits apply per API key; if you regularly need more, talk to us about a custom quota.
What does the API return for a title it does not recognise?+
A valid response with management_level set to null and empty function / standard_roles arrays. You always get the full response shape, so unrecognised titles are easy to detect and handle.
How fast is it?+
Typical latency is single-digit milliseconds per request, with infrequent cold starts adding ~300–600 ms. Even a full 500-title batch completes in well under a second.
Which languages can I integrate from?+
Any language that can make an HTTPS POST request with a JSON body. The docs include ready-to-run examples for curl, PowerShell, and Python.
Need the implementation detail?
Explore request shapes, response fields, and ready-to-run examples.