The same Hermes agents you run in the launcher, on a small Linux machine of their own. Awake when you need them, asleep when you don't, metered by the hour and free while they sleep.
Open the launcher's New agent form and pick cloud as the runtime, or click Create in the web dashboard. Same six questions: which agent, which model, how it signs in, which skills, what the job is.
A machine is provisioned in seconds and goes straight to sleep. Sleeping agents cost nothing but their few GB of storage.
Chat from the launcher window on your laptop, or from the browser console on any device. Same conversation, same tmux session, wherever you pick it up.
Give it a job and an unattended run. It wakes, works, reports, and goes back to sleep. Pro and Studio add scheduled jobs and webhooks.
Your API keys are stored encrypted on our side and written into the machine in its first minute. They are never uploaded from your laptop as files and never shown back to you.
Every lifecycle step, blocker, token count and dollar shows up in the launcher's dashboard and in the web app, the same way local agents report.
$ omarchy-agent-launcher cloud status name state awake h storage last event researcher sleeping 0.23 1.1 GB job done · 2 h ago ops-oncall awake 3.80 0.6 GB heartbeat · 10 s ago rix sleeping 1.05 0.4 GB brief · yesterday $ omarchy-agent-launcher chat ops-oncall → attaching console… ops-oncall: Disk on nas-01 is at 91%. Want me to rotate the old snapshots? I'd free ~40 GB. [blocker: needs your yes] you: yes, keep the last 7 ops-oncall: Done. 38.4 GB freed. Back to watching. $ omarchy-agent-launcher cloud sleep ops-oncall ✓ sleeping · 3.8 awake h metered · $0.00 (inside quota)
Rix is your AI orchestrator, the way J.A.R.V.I.S.* is to Tony Stark. The launcher ships with Rix: a Hermes agent with a SOUL that knows every launcher command. Host Rix here and it can brief you, take delegations, and hand work to your other cloud agents while your laptop is off. Ask "what happened overnight?" from your phone and get the answer.
Sign in with your own account or API key for Anthropic, OpenAI, Gemini, DeepSeek, OpenRouter, xAI and friends, the same list as the launcher. Or point the agent at your private GPU machine and keep every token on a machine only you can reach.
An agent home is a shell, a model client and some files. It does not need a big box. Pick a size per agent; larger sizes unlock with Plus and Pro.
1 vCPU · 1 GB RAM · 5 GB disk
The default. Fine for chat, scripts, research, monitoring.
Fan and up2 vCPU · 2 GB RAM · 10 GB disk
Builds, test suites, a browser for the agent.
Plus and up4 vCPU · 8 GB RAM · 20 GB disk
Big repos, data crunching, several tools at once.
Pro and upEvery size is billed the same $0.10 per awake hour past your quota, and $0.05 per GB-month past your storage.
Hermes Agent today. OpenClaw is next. Both get their own isolated home on the machine, exactly like the launcher does locally.
You get a browser console and a one-shot command runner from the dashboard and the CLI. Full shell access is on the roadmap; today the agent's home is yours to inspect from the console.
The machine reports a heartbeat. After a few idle minutes it goes to sleep and the meter stops. Open the console and it wakes in a couple of seconds.
A hosted agent is a small machine that sleeps between jobs and bills by the awake hour: right for an assistant you talk to every day. A Sandbox is a whole Omarchy environment, optionally a credential-free copy of your twin: right when a new agent or digital employee needs your tools and room to experiment.
Yes. Checkpoint it, download the home, and launch it locally with the same launcher. Nothing about the home is cloud-specific.
* J.A.R.V.I.S. and Tony Stark are trademarks of Marvel. Omarchy.Fans is not affiliated with or endorsed by Marvel.