Hermes Agent can now route inference through Vercel AI Gateway and execute commands inside Vercel Sandbox’s isolated microVMs — no separate provider accounts or API keys to manage.
The integration adds two capabilities Hermes previously lacked:
- 200+ models through AI Gateway at zero token markup
- Unified spend tracking — every request surfaces in the AI Gateway dashboard alongside existing usage
- Isolated execution — agent commands run in ephemeral Linux microVMs with resource limits and automatic timeouts
- Live model picker — the setup wizard pulls real-time availability and pricing from AI Gateway
- Single OIDC token authenticates both Sandbox and AI Gateway
Setup is a picker flow. Install Hermes, select Vercel AI Gateway in the setup wizard, and the picker handles model selection. The full walkthrough lives in Vercel’s changelog.
Sandbox is the piece that makes this safe for agent workflows. Hermes runs arbitrary code — that code spins up in an ephemeral microVM, executes, and disappears. Untrusted commands never touch production systems.
NOTE
Vercel Sandbox ships as a standalone SDK callable from any environment, not just Hermes. If you’re building custom agent infrastructure, the same microVM isolation applies.
Available now. Install Hermes, pick Vercel AI Gateway in the setup wizard — the picker pulls live model availability and pricing automatically.