Data handling
Describes the code as built. Not yet deployed.
On this page
What do you log?
Never your input. For each answered call: the decision name, the answer type, the abstain flag, the model, whether a rule settled it and how long the model took. For a failed call: the error’s type and status.
What do you record for billing?
For each call made with an API key or a connected app’s token, one usage row: the account, the key, the channel, the decision name, the answer type, the input-token count, the cost, the abstain flag, the status and how long the model took. Never your input.
Are responses cached?
No. Every response, errors included, is marked cache-control: no-store.
What do you keep for an Idempotency-Key?
Only if you send one: the key, a SHA-256 hash of the request and the successful response, for 24 hours, so a repeat gets the same answer. Not the request itself. The response can include a transaction id you sent, as the answer to a match.
Does the model provider keep or train on our data?
By default, the API requests routing only to model providers with zero-data-retention and no-training agreements; that isn’t yet verified by a live call. We’ll confirm the details in writing before your first live call.
What shouldn’t we send?
- Names, account numbers or balances.
- Health readings or protected health information, including in your own questions.
Card numbers, Social Security numbers, and account numbers with their routing numbers are refused: a 400 with code: "sensitive_data" that names the field, before any model sees the input. The value isn’t repeated or kept.
SOC 2, a BAA or data residency?
We make no claims about SOC 2 reports, business associate agreements or data residency. If you need them, say so in the form and we’ll answer in writing.
Health decisions?
The three health decisions are built and tested against a stand-in model, and none is open. We make no claim about HIPAA or a business associate agreement. Do not send protected health information.
# Data handling Describes the code as built. Not yet deployed. ## What do you log? Never your input. For each answered call: the decision name, the answer type, the abstain flag, the model, whether a rule settled it and how long the model took. For a failed call: the error’s type and status. ## What do you record for billing? For each call made with an API key or a connected app’s token, one usage row: the account, the key, the channel, the decision name, the answer type, the input-token count, the cost, the abstain flag, the status and how long the model took. Never your input. ## Are responses cached? No. Every response, errors included, is marked `cache-control: no-store`. ## What do you keep for an Idempotency-Key? Only if you send one: the key, a SHA-256 hash of the request and the successful response, for 24 hours, so a repeat gets the same answer. Not the request itself. The response can include a transaction `id` you sent, as the answer to a match. ## Does the model provider keep or train on our data? By default, the API requests routing only to model providers with zero-data-retention and no-training agreements; that isn’t yet verified by a live call. We’ll confirm the details in writing before your first live call. ## What shouldn’t we send? - Names, account numbers or balances. - Health readings or protected health information, including in your own questions. Card numbers, Social Security numbers, and account numbers with their routing numbers are refused: a 400 with `code: "sensitive_data"` that names the field, before any model sees the input. The value isn’t repeated or kept. ## SOC 2, a BAA or data residency? We make no claims about SOC 2 reports, business associate agreements or data residency. If you need them, say so in the form and we’ll answer in writing. ## Health decisions? The three health decisions are built and tested against a stand-in model, and none is open. We make no claim about HIPAA or a business associate agreement. Do not send protected health information. Source: /api/docs/data