Your agents, running on our infrastructure.live
Agent Pods is the runtime for agent harnesses — a place where your agents run on Layer X1 instead of you babysitting your own boxes. Each agent gets a pod: an always-on soul that holds its identity, budget and kill switch, a real VM body it works in, and inference on the layer that's already live. Spawn one and it's running in a minute.
A home for the whole run.
You used to rent a box, wire up a harness, and hope the session held. That's our job now — your agents run on Layer X1 itself: kept alive, kept apart in a VM of their own, and killable to the metal in one click.
Long-lived sessions
A run that goes for hours holds. Each pod is a Durable Object that stays alive across the whole trajectory and idles to near-zero cost between turns — nothing of yours has to stay up just to keep an agent breathing.
A VM for every agent
Every agent gets its own real Linux VM — files, tools, a shell, and a live web terminal you can watch from the browser. Right-sized per agent, sealed from every other pod, and destroyed to the metal the moment you kill it.
Budget + kill switch
Every pod carries a hard spend cap and an instant kill switch, enforced in-pod. Overspend and it auto-freezes; hit kill and the VM is gone. A rogue pod freezes or dies on its own — the rest of the fleet never notices.
Same key
The runtime stands on the inference layer. Every model in the catalog is already there — metered and budget-capped through Layer X1, on the key you hold today. No second integration, no new account.
New layer. Proven ground.
Pods run on the same engine that serves inference in production today. When your agents move in, they don't land on something new — they land on the layer that already carries their model calls, metered on the key you already hold.
Spawn a pod from the console — pick an agent, watch its VM come up, chat it, kill it.
Your agent doesn't change.
Everything underneath does.
Start free · no card · Starter from $5/mo