How Vercel Agent accesses production safely

Vercel Agent addresses the challenge of granting agents broad production access by proposing a structured workflow. Instead of requesting immediate, unrestricted permissions, the agent first investigates the issue and presents a detailed plan outlining the actions it will take and the specific access it requires. This allows users to review the proposed solution before authorizing any changes, ensuring transparency and control over production environments.

The system enforces access through scopes, which are named capabilities recognized by Vercel’s authorization framework. Plans that request access beyond a read-only baseline must be approved as a whole, including any read-only tasks they contain. For example, a broken deployment might trigger an investigation under a read-only plan, followed by a redeploy plan that requests additional write access, both of which require user approval.

Vercel Agent operates as a Vercel App, authorized by the user within a team context. The authorization is limited by the user’s existing permissions, and the OAuth service exchanges this for shorter-lived tokens with specific scopes. These tokens define the exact capabilities permitted, such as reading deployments or modifying project settings, and are enforced by Vercel APIs on each request.

The system ensures that plans are validated against the latest scope definitions to prevent mismatches between requested and available permissions. For instance, if a plan requests an outdated scope, the system rejects it and returns valid alternatives, requiring an explicit correction before approval. This approach prevents fallback to broader access and maintains strict enforcement of permissions throughout the agent’s operations.