A box per project
Every project gets its own isolated machine. Separate agents, separate tools, separate blast radius.
Deploy any AI agent
in 60 seconds.
cloudrun provisions an isolated box per project, configures the runtime, and hands you one canvas over the fleet.
Connectors
Agents reach the tools you already run
Connect a service once, at the agent level. The connection is held for you, and no credentials get pasted into a prompt.
Control plane
A control plane, not another runtime
Your agents run on their own box. cloudrun stands it up, configures it, and gives you one place to watch the whole fleet.
One canvas over the fleet
Organization, project, engine, agent. Every node on one canvas you arrange, instead of a dashboard per agent.
Your keys, your models
Agents start on the runtime's free tier. Connect your own providers when you want stronger ones.
Reachable where you work
Telegram, Slack, email and more. A channel is configuration on the agent, not a rewrite.
Automations on a cron
Scheduled prompts run on their own timer. The agent wakes up, does the work, and reports back.
Paused, then reclaimed
A lapsed subscription pauses the fleet and keeps it for 30 days before anything is deleted.
Provisioning
Sixty seconds, not thirty minutes
Boxes do not install a toolchain at provision time. A golden snapshot is baked ahead of the request, and standing up a project restores it.
Baked, not built
The runtime, its dependencies and a browser engine are already in the image. Provisioning is a restore.
Charged before it boots
The seat is reserved first, so a declined card can never leave a running box behind.
Paused, then reclaimed
A lapsed subscription pauses the fleet and keeps it for 30 days before anything is deleted.
- Project createdseat reserved before anything boots
- Box requestedisolated machine, one per project
- Snapshot restoredruntime and browser already baked in
- Runtime configureddriven over its own dashboard API
- Agent onlinecanvas ready, free model tier live
Engines
One contract, room for every runtime
An engine kind brings its own substrate, its own panels and its own API namespace. There is no plugin registry to satisfy, so adding the next one is wiring, not a rewrite.
Hermes
Profiles with a soul, skills, tools, channels, automations and memory. Available now.
OpenClaw
Wired as an engine kind. Coming soon.
Codex
Wired as an engine kind. Coming soon.
Claude Code
Wired as an engine kind. Coming soon.
OpenCode
Wired as an engine kind. Coming soon.
Pi
Wired as an engine kind. Coming soon.
Cursor
Wired as an engine kind. Coming soon.
Pricing
You pay for running deployments
Not for members, not for projects, not for a tier you grew out of. The subscription quantity always equals the number of deployments actually running.
Cloud
An isolated box per project
Hermes configured and online in about a minute
Free model tier to start, your own keys when you want more
Channels, automations and memory on the agent
One canvas over the whole fleet
How the meter works
Per deploymentMembers are never billed, only running deployments
Delete a deployment and the seat count drops with it
A lapsed subscription pauses the fleet, and keeps it for 30 days
What an agent gives back
Set the work a person does by hand today and how much of it an agent can take. The numbers update as you type.
Net savings per year
$38,292
$3,191 a month once the $49 deployment is paid for, with 72 hours handed off.
min
$
60%
The work today$5,400
With agents$2,209
Questions, answered
What the control plane does, what it does not, and what you are charged for. Anything missing is in the docs.
No. It is a control plane, not a runtime. Your agents run on their own box, and their skills, channels, automations and transcripts live there. cloudrun provisions the box, configures it, and stores only what it needs to reach it.
Your first agent is one project away
Create a project, pick Hermes, and the box is yours in about a minute.
~60 s
To a live agent
1
Box per project
7
Engine kinds wired
30 days
Fleet kept after a lapse