What it covers
The decision endpoints are/v1/decide, /v1/systemone and /v1/hooks/claude-code. Zero data retention for inference covers the whole organisation: every API key, every member and every console session.
What is kept
Usage records stay: for each request, when it was made, who made it and with which API key, the model, the tokens, the cost and how it ended.What is out of scope
Fine-tuning is out of scope: datasets, labels, evaluations, prompt tuning and fine-tuned models. Zero data retention for inference doesn’t change how they are kept (data residency).How to get it
We switch zero data retention for inference on for an organisation once it has signed a ZDR addendum with us. To start, contact support at help@sqwish.ai.Switch it off and on
Once we have switched it on, the organisation’s owners can switch it off and on again in Organisation settings in the console. The console asks first. Every owner gets an email 10 minutes after a change, once it has taken effect. Changes one person makes close together come in one email. If we withdraw it while it is on, it stays on until an owner switches it off. Your code can check it too:GET /v1/account returns user.zero_data_retention_since, when it was switched on, or null while it is off.
What changes for retries
Without zero data retention, a decision sent without anIdempotency-Key keeps no answer at all. One sent with a key keeps its answer, encrypted, for 24 hours. A retry with the same key and body gets that answer back, without running or charging again (supported idempotency).
With zero data retention, no answer is kept, so an Idempotency-Key isn’t honoured. A retry with the same key runs the request again and is charged again. Our SDKs retry by themselves after a network error, so that retry is charged again too (retries and idempotency keys).