Runner flavors: PowerShell, C#, Python, and Agent

CloudRadial AutomationAI runners come in four flavors — PowerShell, C#, Python, and Agent — and a workflow's steps are routed to the matching flavor automatically. This article explains each flavor, when to use it, and how routing works. It is for Admin and Owner roles.

• The Four Flavors
• How a Node's Language Routes to a Runner
• The Agent Flavor

The Four Flavors

Each flavor is a distinct runtime, because an Azure Functions app hosts one runtime per app:

• PowerShell — runs PowerShell Script nodes; use it for Microsoft and Windows administration and most MSP tooling
• C# — runs C# Script nodes; use it when you want typed .NET code
• Python — runs Python Script nodes; use it for Python libraries and data work
• Agent — runs Agent nodes, the goal-directed AI flavor described below

When you install a runner, all four flavors are deployed together so any node type can be dispatched. You do not pick a flavor when registering a runner — you register one runner and it covers all four.

How a Node's Language Routes to a Runner

Every step a run produces is tagged with the language of its node. A runner flavor only ever sees steps of its own language: a PowerShell runner leases only PowerShell steps, a Python runner only Python steps, and so on. You do not route work by hand — AutomationAI matches each step to the correct flavor automatically. As a result, a workflow that mixes a PowerShell node, a Python node, and an Agent node has each step picked up by the matching flavor of the deployment's runner.

The Agent Flavor

The Agent flavor is different from the three script flavors. Where a script flavor leases and executes a single script step, the Agent flavor runs goal-directed, multi-turn work: it carries out an Agent node's goal using the tools your installed extensions provide, taking several turns of reasoning and tool calls. It is the same execution model that powers Ask AI. Because it calls AI models, Agent work consumes Usage Credits. For authoring detail on the node itself, see the Agent node article in the Node reference group.