Skip to main content
Most runs need a machine: that is where files are read and commands run. But many useful automations do neither. “When a customer emails, label it and post to Slack” talks only to Gmail and Slack. A web-research digest only reads web pages. These can run with no machine, entirely on Reliant’s servers, so they fire even when your laptop is shut and no cloud machine is awake. You choose this when you activate an automation, by picking No machine in its Runs on control (schedules created with reliant trigger create always need a machine).

​
What a no-machine run can use

A no-machine run is offered only tools that run server-side: web access, integrations, and planning tools. It is not offered the shell, file editing or anything else that needs a machine.
An agent works too, if you limit its tools to ones that need no machine:
The stock agent’s default tools include the shell and file tools, so leaving tools at its default makes it a machine workflow. Setting it to ["tag:web"], as above, does not.

​
What needs a machine

Reliant checks this before the automation is saved, and the workflow builder warns you as you author. These make a workflow need a machine:
  • a run node (a shell command);
  • a create_worktree node;
  • an action that runs on a machine;
  • an invoke_tool of a machine tool;
  • a daemon: selector on the workflow or a node;
  • an agent that is given machine tools in its tools list.
If you try to turn on No machine for a workflow like this, the app refuses and names what needs one. Activating a no-machine automation whose agent is given machine tools is refused for the same reason: dropping those tools silently would quietly change what an unattended automation does.