> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rayline.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Data and privacy

> When ARC uses requests to improve, and how to choose.

Each organization has a data setting, shown under **Organization** > **Data use** in the platform:

| Setting | What it means |
| - | - |
| Standard (default) | Requests aren't used for training. Rayline may retain hosted generation content for up to 30 days for abuse prevention, security investigations and support. Inference providers must not train on requests and may retain them for no more than 30 days. |
| Contributor | Eligible hosted generation requests may be used to improve ARC routing. Owners control access with **Allow Contributor models**. |
| Zero data retention | Rayline and the inference endpoints used for the request retain no request content. Sales approval is required before use. |

Owners can switch between Standard and Contributor themselves, subject to any contractual settings. All zero data retention use requires sales approval, including requests that name a ZDR model ID. Use **Talk to sales** under **Organization** > **Data use** to arrange access.

On the Rayline API, an ARC request follows its model ID rather than this setting, as described below. [Decisions requests](#decisions-api) have their own content-retention behavior. See [Rayline's privacy policy](https://rayline.ai/privacy) for the full policy, including usage records and legal retention requirements.

## ARC model IDs choose the data mode

On the Rayline API, the ARC model ID you send decides the data mode of an ARC request, not the organization's setting:

| Model IDs | Data mode |
| - | - |
| `arc-swift-1.0`, `arc-sage-1.0`, `arc-1.0` | Standard |
| `arc-swift-contributor-1.0`, `arc-sage-contributor-1.0`, `arc-contributor-1.0` | Contributor |
| `arc-swift-zdr-1.0`, `arc-sage-zdr-1.0`, `arc-zdr-1.0` | Zero data retention |

The [Decisions API](https://docs.rayline.ai/arc/decisions-api.md) uses its own IDs with the same data modes: `arc-decisions-0.1` (Standard), `arc-decisions-contributor-0.1` (Contributor) and `arc-decisions-zdr-0.1` (zero data retention). Use any ZDR ID only after sales has approved your access.

An organization on zero data retention can use only the zero data retention IDs. Any other ARC ID is refused with a `403` `zdr_model_required` error, which names the ID to use, before any model is called. Requests to other models follow the organization's setting.

| Surface | Used to improve ARC | Exploration |
| - | - | - |
| ARC Swift, ARC Sage, ARC Dynamic (hosted inference) | Only with a Contributor model ID | About 5% of requests with a Contributor model ID; 0 otherwise |
| Routes API | Never | None by default; only the routing temperature you set |
| Decisions API | No, in any data mode; see [below](#decisions-api) | None |
| Self-hosted routing | Not covered by this hosted policy | Set by your own configuration |

A local connector that forwards to hosted inference follows the hosted rows.

When Rayline cannot confirm your organization's data setting for a request, it handles the request as zero data retention: no exploration, no training use, and only the zero data retention ARC IDs are served.

## Decisions API

ARC reads a Decisions request's state and questions to answer it, and keeps neither once it has answered, in every data mode. Rayline's logs and usage records for a Decisions request carry token counts, timings and metadata, never your state or question text.

## Contributor models

A request to `arc-swift-contributor-1.0`, `arc-sage-contributor-1.0` or `arc-contributor-1.0` runs in Contributor, whether your organization's setting is Standard or Contributor and whatever your own privacy setting.

* Organizations on zero data retention can't use Contributor models: the request is refused with `zdr_model_required`.
* Owners can turn Contributor models off for the organization with **Allow Contributor models** under **Organization** > **Data use**. A request to a Contributor model is then refused with a `403` `contributor_models_off` error before it reaches any model.
* Your usage lists these requests under the Contributor model you named.

A model that Standard or zero data retention doesn't allow is refused with a `privacy_mode_model_unavailable` error rather than swapped for another model. Choose another model, or switch the organization to Contributor.

The organization-level data setting is rolling out. Until it is available for your organization, contact [hello@rayline.ai](mailto:hello@rayline.ai) to confirm how your requests are handled. Workshop data is governed by [Workshop's terms](https://docs.rayline.ai/workshop.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.