Vercel AI Gateway and Vercel Sandbox now available on Hermes Agent.

Hermes Agent users can now route model requests through Vercel AI Gateway and execute agent commands inside isolated Vercel Sandbox microVMs. The integration, announced in Vercel’s changelog, gives developers a more centralized way to manage model access, spending and potentially risky command execution.

The update connects two important parts of Vercel’s agent stack directly to Hermes Agent, the open source agent from Nous Research. AI Gateway becomes the agent’s inference layer, providing access to more than 200 models. Sandbox, meanwhile, supplies an isolated runtime for the commands an agent runs on a developer’s behalf.

It is a practical combination. Model routing and code execution are separate concerns, but developers building autonomous agents usually need to manage both.

When configured with Vercel AI Gateway, Hermes sends its inference requests through Vercel rather than requiring each model connection to be managed independently. Vercel says users can access more than 200 models through the gateway with no markup on model tokens.

Every request also appears in the AI Gateway dashboard alongside the user’s other usage and spending. That consolidated view should be useful for small teams experimenting with several models, particularly when an agent may make many calls while completing a single task. Instead of reviewing activity across multiple provider dashboards, builders can monitor the relevant requests in one place.

The gateway layer also gives developers access to proxying, caching and rate-limiting capabilities for AI API calls. Those controls can matter once an internal experiment becomes a customer-facing workflow, where duplicated requests or an agent caught in a loop can create unexpected costs.

The Hermes integration arrives as Vercel continues adding models and agent tooling around AI Gateway. The company has also added Grok’s image generation model, as covered in Vercel Adds Grok Imagine Image 2.0 to AI Gateway, broadening the range of workloads available through the same inference layer.

The Sandbox integration addresses a different problem. Agents that can inspect files, install packages or run terminal commands are more capable than chat-only assistants, but they also create a larger security risk. A mistaken or malicious instruction could modify the host environment, expose data or leave unwanted processes running.

Hermes can now execute each agent command inside a Vercel Sandbox microVM. Isolation does not eliminate the need for careful permissions and review, but it gives the command a dedicated boundary rather than running it directly in the developer’s primary environment.

For builders, the two integrations divide the workflow cleanly:

• AI Gateway handles model inference across more than 200 models and records requests, usage and spending.
• Vercel Sandbox runs agent commands in isolated microVMs.
• Hermes Agent remains the interface coordinating model decisions and terminal actions.

This separation is especially relevant for coding agents, research automations and operational assistants that need to move beyond generating text. It also fits Vercel’s broader effort to package reusable agent infrastructure, following additions such as shareable Skill packs.

New users can install Hermes with its published shell installer, then select Vercel AI Gateway during the setup wizard and paste in their gateway key. They can also choose Vercel Sandbox as the terminal environment, causing agent commands to run in microVMs.

Existing Hermes installations need to be updated first. Users can then run hermes setup model to select Vercel AI Gateway and hermes setup terminal to choose Vercel Sandbox. Vercel recommends running hermes doctor afterward to verify the configuration.

The integration is modular, so existing users can switch the inference layer, the command runtime or both. That makes adoption less disruptive for teams that want centralized model access but are not ready to change execution environments, or those primarily interested in isolating terminal activity.

What is Vercel AI Gateway and Vercel Sandbox now available on Hermes Agent?

It is an integration that lets Hermes Agent use Vercel AI Gateway for model inference and Vercel Sandbox for isolated command execution. AI Gateway offers access to more than 200 models, while Sandbox runs agent commands inside microVMs.

How can existing Hermes Agent users enable the integrations?

Users should update Hermes, then run hermes setup model to choose Vercel AI Gateway and hermes setup terminal to choose Vercel Sandbox. The hermes doctor command can verify that the resulting setup is working.

Does Vercel add a markup to model token prices?

Vercel says AI Gateway provides access to more than 200 models with no markup on tokens. The announcement does not provide additional pricing or plan requirements for Gateway or Sandbox usage.