Vercel Adds Devin Outposts Support to Isolated Sandbox MicroVMs.

Vercel is giving Devin’s AI coding agent a managed place to run. The company announced support for Devin Outposts, allowing every Devin session to execute inside its own isolated Vercel Sandbox microVM instead of relying on a local Outpost worker that must remain online.

The integration divides responsibility between the two platforms. Devin’s control plane remains with Cognition, where its agent loop handles inference and planning. Session orchestration and command execution happen inside the customer’s Vercel project, while a Vercel durable workflow manages the lifecycle of each session.

That distinction matters for teams evaluating where AI-generated commands actually run. Devin is not being fully self-hosted on Vercel, but its working environment and command execution move into an isolated, project-controlled sandbox.

Each Devin session receives a separate Sandbox microVM. This provides an execution boundary between sessions and removes the operational requirement to maintain a local Outpost worker, which would otherwise need to stay connected and available.

Vercel also connects the integration to two existing Sandbox controls:

• Snapshots: Teams can use Sandbox snapshots to preserve session state. This can help an agent resume work without rebuilding its entire environment from scratch.
• Firewall rules: Devin network policies can be implemented through Sandbox firewall rules, giving teams control over the external services and domains a session can reach.

For small development teams, the practical benefit is less infrastructure to operate around an AI agent. A developer can start a Devin session, assign it a Vercel-hosted environment and let the platform manage the underlying session lifecycle. The firewall integration is particularly relevant when agents need access to package registries or APIs but should not receive unrestricted outbound connectivity.

The release also expands Vercel’s positioning of Sandbox as execution infrastructure for coding agents, rather than only a utility for running untrusted code. It follows the addition of Sandbox support for another agent environment, as Create With recently covered in the Hermes Agent integration.

Setup begins with Vercel’s Devin Outpost quickstart. After deploying it, a Devin administrator must approve the connection.

The integration then adds encrypted credentials to the Vercel project automatically. Vercel says administrators do not need to manually copy a token or API key, reducing one common source of configuration mistakes and credential exposure.

Once connected, a user starts a session in Devin and chooses the new Vercel outpost as the virtual environment. Commands from that session execute in its assigned Sandbox microVM, while the durable workflow coordinates its lifecycle. Teams can then layer snapshots and firewall policies onto the environment according to their recovery and access requirements.

Vercel says the integration is available to Pro and Enterprise teams. The announcement does not specify separate pricing for Devin Outposts, so prospective users will need to evaluate the requirements of both their Vercel plan and their Devin deployment before adopting it.

What is Vercel Devin Outposts?

Vercel Devin Outposts is an integration that runs each Devin coding session inside an isolated Vercel Sandbox microVM. Devin continues to handle inference and planning through Cognition’s control plane, while orchestration and command execution run within the customer’s Vercel project.

Which Vercel plans support Devin Outposts?

Vercel says the integration is available to Pro and Enterprise teams. The announcement does not provide separate pricing details or describe additional usage charges.

How is the integration installed?

A team deploys Vercel’s quickstart and has a Devin administrator approve the connection. Encrypted credentials are added to the Vercel project automatically; users can then select the Vercel outpost as the virtual environment when starting a Devin session.