We heard Ship AI, and immediately switched into “developers, developers, developers” mode.
AI Gateway
One line of code to switch models. Switch from Claude to GPT, Gemini, or Groq without any rewrites, no SDK graveyards, no crying in prod.
Unified billing, better rate limits, and observability so you actually know what went wrong this time.
Workflows
Your agent can think, wait, retry, pause for a week, then wake up like nothing happened. No cron jobs, no queues held together with hope, no spaghetti infrastructure.
Sandboxes
LLMs can write code. LLMs can also break everything. So Vercel made isolated micro-VMs so your agent can “experiment” without deleting prod.
We appreciate that kind of babysitting.
Fluid Compute
Only pay when the CPU actually works. If your agent is sleeping, thinking, or doing whatever agents do… you don’t pay for silence.
Marketplace
Install agents instead of building them from scratch. Do you need debugging, anomaly detection, or a code reviewer that never sleeps?
Add, deploy, done.
Check out the full breakdown (and find out how many times we almost broke prod)
