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

# Secrets management

> Where Orq.ai stores API Keys, provider credentials, secret variables, and signing secrets, who can read each one, and how to rotate it.

**Orq.ai** stores, exposes, and rotates each kind of secret differently. The table below maps each one and links to the page that owns it.

## Where each secret lives

| Secret                                                                                                                                | Stored as                           | Readable by                                        | Rotation                                       |
| ------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------- | -------------------------------------------------- | ---------------------------------------------- |
| [**API Keys**](/docs/ai-studio/organization/api-keys) (`sk-orq-…`)                                                                    | SHA-256 hash and displayable prefix | Create response only                               | Create a replacement, then revoke the old      |
| [**Management Keys**](/docs/ai-studio/organization/management-keys)                                                                   | SHA-256 hash and displayable prefix | Create response only; admin-only page              | Create a replacement, then revoke the old      |
| [Provider keys](/docs/ai-gateway/providers-overview) (BYOK)                                                                           | Encrypted with a workspace key      | Never returned; masked on read                     | Update the key on the provider panel           |
| [Secret variables](/docs/ai-studio/ai-engineering/run-agents) in runs and [schedules](/docs/ai-studio/ai-engineering/schedule-agents) | Plain text in the schedule payload  | Anyone who can read the schedule                   | Store the payload again                        |
| Notifier webhook headers                                                                                                              | Plain text                          | Returned empty only when marked `{"secret": true}` | Update the header                              |
| [Webhook signing secret](/docs/ai-studio/organization/webhooks)                                                                       | Plain text                          | Create response only; masked on read               | Generate a new secret, then update the webhook |
| [MCP server credentials](/docs/ai-gateway/mcp-portal/mcp-servers)                                                                     | Encrypted with a workspace key      | Never returned; masked preview                     | Update in the server settings                  |
| [SSO credentials](/docs/ai-studio/organization/sso)                                                                                   | Encrypted with a workspace key      | Never returned                                     | Re-enter the credential                        |
| [External knowledge base key](/docs/ai-studio/ai-engineering/external-knowledge-bases)                                                | Encrypted with a workspace key      | Never returned                                     | Update the key                                 |

## Keeping secret variables out of traces

Mark a variable as secret with `{"secret": true, "value": "..."}`, as described in [Run agents](/docs/ai-studio/ai-engineering/run-agents). **Orq.ai** substitutes the value into instructions and tool headers, then redacts it wherever the run is recorded:

* Traces and span attributes: the value is replaced with `***`.
* Stored response variables: the secret key is removed.
* The live response: still contains the rendered value.

<Warning>
  A secret variable redacts its value at run time only. A value stored in a schedule payload, a notifier header, or a webhook definition stays readable wherever that resource is readable.
</Warning>

For content-level masking of request and response bodies, see [Mask sensitive content in traces](/docs/ai-gateway/features/security).

## Encryption at rest

* **Encrypted with a workspace key** (AES-256-GCM): provider, MCP server, SSO, and external knowledge base credentials. A database read alone does not disclose them.
* **Hashed**: API Keys and Management Keys store a SHA-256 hash and a displayable prefix, never the token.
* **Plain text**: schedule payload variables, notifier headers, and webhook signing secrets. Schedule payload variables rely on access control alone; notifier headers are masked on read only when marked `{"secret": true}`, and webhook signing secrets are always masked on read.