aimldocs
Console

Data policy

Decide what is archived and for how long, redact secrets before archiving, turn on zero data retention, and limit providers and regions.

The data policy decides what happens to your prompts and responses: whether they are archived, for how long, which providers may see them, and where they are processed. Open it from the Data policy tab under Settings.

The settings here are the defaults for the whole organization. Projects and keys can only be stricter, never looser. See projects and API keys.

Changing the policy needs the owner or admin role. Other roles can read it.

The page has two parts: the policy form and the Per-provider policy table.

Logging and retention

Archive prompts and responses turns the payload archive on or off. With it on, you can open a request later and read what was sent and returned. See requests.

Retention is how long an archived payload is kept: 7, 30, 90 or 365 days. When that time is up, the payload is made unreadable for good.

Images and audio are never archived.

Redaction before archive

Redaction removes sensitive values from the archived copy. It never changes what reaches the provider. A removed value is replaced by a marker that says which rule matched.

Tick the presets you want:

  • API keys: keys and tokens from common providers and services, and bearer tokens.
  • Card numbers: card numbers that pass the standard check digit test.
  • Aadhaar: Aadhaar numbers.
  • PAN: Indian income-tax PANs.
  • Email addresses
  • Phone numbers: Indian and international formats.

Hover a preset to see what it catches.

Custom patterns

Under Custom patterns (one regex per line), add your own regular expressions, one per line. A policy can hold up to 20 patterns. A pattern is refused if it is empty, longer than 200 characters, or able to match an empty string, because such a pattern would redact everything. An invalid pattern shows its error under the field.

Preview

Preview on a sample shows what your presets and patterns would do. Type or paste some text, or leave the field empty to use the built-in sample. The box below updates as you type and shows the number of redactions and the redacted text.

Use the preview before you save. Nothing in it is stored.

Zero data retention

Zero-data-retention mode is one switch that does four things:

  • Nothing is archived, so the archive checkbox and the retention selector are disabled.
  • The response cache and request replay are off.
  • Requests are routed only to endpoints that offer zero data retention.
  • The end-user ids you send are kept only as a hash.

When you tick it, the page checks which models would stay available and lists the ones that would not. Press Save data policy and a confirmation appears with the number of models that become unavailable. Press Confirm and save to turn it on.

While it is on, the page reminds you that routing only to these endpoints can reduce capacity during a provider incident.

Providers and regions

No-train: exclude providers that train on API data removes every provider whose terms allow training on what you send.

Allowed processing regions (none = any) limits where your requests are processed: India, United States or European Union. Tick none to allow any region. If you allow India only, models hosted only elsewhere become unavailable.

When you have ticked some regions but not all, a second row appears: If nothing in those regions can serve a request, it may be served from. Tick a region there to prefer your own regions and fall back to it when none of them can serve. A request served from a fallback region carries a residency_fallback warning naming where it was served. Tick none and the policy is strict: the request is refused. A project or key that sets its own regions is strict unless it names its own fallback, and that fallback can only be a region the organization already allows.

A request that no allowed endpoint can serve is refused. It is never sent to a provider your policy excludes.

Save

Press Save data policy. Keys pick up the new policy within a minute.

Per-provider policy

This table shows what each provider does with API data, taken from its published terms:

  • Trains on data: yes or no.
  • Retention: how many days the provider keeps data, or none.
  • ZDR: whether zero data retention is available.
  • Regions: where the provider processes requests.
  • Source: a link to the provider's terms.
  • Verified: when the entry was last checked.

The no-train and zero-data-retention settings above filter on these values.

Creating an organization accepts the terms of service, the acceptable use policy and the data processing agreement for it, and the console does not ask again. We record the version of each document, the time and the address with your account. When a document changes, the new version is announced ahead of time by the legal.updated webhook, and continuing to use the console accepts it.

The list of sub-processors is published at https://api.ai.ml/v1/legal/subprocessors. A change is announced ahead of time, by email and through the subprocessors.changed webhook. To object, restrict the provider in the policy above.

For a copy of the data processing agreement or the DPDP notice template filled in with your organization's details, write to support. The legal routes are in the legal API reference.

On this page