Run your agents where you need them.
The same agent — one manifest, one SDK — runs on your own machine, not just in the cloud. It connects outbound over a secure stream, so it works behind a firewall, and can use models running right on the device.
Three ways to host
Desktop App
An Electron app that hosts agents in isolated processes, runs on-device LLM inference, and manages your device fleet. Sign in and install an agent in a few clicks.
Headless CLI
A terminal daemon for servers and homelab GPU boxes — no screen required. Inference-first: pull a model and serve it to your organization.
Mobile
The companion mobile app grows from monitoring into a host, so an agent can run on your phone.
Outbound-only by design
A device-hosted agent opens an outbound stream to pull work and posts results back — no inbound ports, no public IP. The control plane treats it exactly like a cloud agent; only the host and the delivery channel differ.