Autonomous Agents (also Choria Machines) are implemented as Finite State Machines that you describe in a file machine.yaml.
A state is a named stable state the machine can find itself in. If you are managing the room air quality using an HVAC system you would have states unknown, idle and running.
The machine can receive events, for example in the HVAC example you would periodically check the air quality and should it be considered bad using your own criteria you’d fire the air_bad event that would move the machine into the running state. Likewise when the air is good you’d fire the air_good event to move to the idle state.
Finally it would have a variables_changed event that can fire if the JSON file with your air quality requirements in it change, and a variables_unknown event should the JSON file be missing.
These events - air_good, air_bad, variables_unknown and variables_changed - are called transitions.
In the HVAC example above you would have a 4 watchers.
First you’d have a file watcher to check the JSON file containing your measurements. If the file is missing it would trigger a variables_unknown transition keeping the macine in the unknown state. If the file is there or changes it would fire a variables_changed transition moving the machine to idle state.
Second you’d have a script that can turn the HVAC off and keep it off, this watcher would be valid only in the unknown and idle states and would run regularly. This ensures the HVAC is off when unknown and keeps it off should someone fiddle the switches.
Third you’d have a script that turn the HVAC on and keep it on, this would only be executable in the running state and would be executing regularly there to ensure it stays on.
Finally you’d write a small script that would query your sensors and determine if the air is good or bad. This is an exec watcher that is valid in any of the idle or running states and it would run on a regular interval like for example 1m.
Choria has a standard set of events - alive, startup, shutdown, provisioned and more - state transitions and state announcement will publish their own event types. Dashboards will be made that can passively observe a running network of machines.
A State Machine combines states and transitions to describe every possible situation the managed item might find itself in and how it gets to that situation or state.