Agents
API keys and scopes
The credential model for programmatic access — scopes, restrictions, rotation, and revocation.
Programmatic access runs on scoped API keys issued by the platform's authentication service. A key is a bearer credential that acts as you, within the boundaries you set when creating it.
Scopes
Permissions are distinct and independently grantable:
| Scope | Allows |
|---|---|
| Read | List and read artifacts, datasets, and outputs |
| Write | Create and edit artifacts and their content |
| Render | Request renders and read job status |
| Share | Create and revoke shares |
| Publish | Change publication state with reviewed metadata |
| Delete | Move artifacts to trash and remove them |
Restrictions
- Artifact restriction — optionally pin the key to specific artifacts; requests outside the set fail authorization.
- No credential minting — an agent cannot create or rotate another agent's credentials.
- Ownership first — scopes never bypass ownership checks on another user's resources.
Lifecycle
- Create with least privilege; the key value is revealed exactly once.
- Rotate as routine hygiene or after suspected exposure — rotation replaces the credential in one step.
- Revoke immediately when an integration is retired; the next request with a revoked key fails.
Handling rules
Treat keys like secrets
Keep keys in your secret manager or process environment. Never place them in source code, project files, command-line arguments, URLs, logs, or analytics.
- One key per integration, named for what it belongs to.
- Prefer artifact-restricted keys for agents working on specific artifacts.
- The console view of the same model is documented under API keys.