Flow Studio Node — AI Functions

AI Function Node in Flow Studio

Drag the "AI Function" node from the node palette onto the workflow canvas. Configure it by selecting an AI Function from the dropdown and mapping input/output variables:

// AI Function Node configuration (stored as workflow node definition)
{
  "nodeType":     "AIFunction",
  "functionName": "calculate_leave_days",
  "inputMapping": {
    "startDate":    "workflow.leaveStartDate",
    "endDate":      "workflow.leaveEndDate",
    "annualEntitlement": "workflow.employee.leaveEntitlement",
    "usedDays":     "workflow.employee.leaveUsed"
  },
  "outputMapping": {
    "approved":      "workflow.leaveApproved",
    "remainingDays": "workflow.daysAfterLeave",
    "message":       "workflow.approvalMessage"
  }
}

AIFunctionNodeExecutor

// Flow Studio execution node for AI Functions
public class AIFunctionNodeExecutor : BaseNodeExecutor<AIFunctionNodeSettings>
{
    protected override async Task<NodeExecutionResult> ExecuteInternalAsync(
        NodeDataContext context,
        CancellationToken ct)
    {
        var settings = GetSettings(context);

        // Resolve the named function from the store
        var function = await _store.GetByNameAsync(
            settings.FunctionName,
            context.TenantId, ct)
            ?? throw new ConfigurationException($"AI Function '{settings.FunctionName}' not found");

        // Build input object from workflow variables
        var inputDict = new Dictionary<string, object?>();
        foreach (var (funcParam, workflowVar) in settings.InputMapping)
        {
            inputDict[funcParam] = context.GetVariable(workflowVar);
        }
        var inputJson = JsonSerializer.SerializeToElement(inputDict);

        // Execute the function
        var result = await _runtime.ExecuteAsync(function, inputJson, new AIFunctionContext
        {
            TenantId = context.TenantId,
            AgentId  = context.AgentId ?? Guid.Empty,
            UserId   = context.UserId ?? string.Empty
        }, ct);

        // Map outputs back to workflow variables
        foreach (var (funcOutput, workflowVar) in settings.OutputMapping)
        {
            if (result.TryGetProperty(funcOutput, out var value))
                context.SetVariable(workflowVar, value);
        }

        return NodeExecutionResult.Success();
    }
}

Comparison: AI Function Node vs. AI Agent Node

Aspect	AI Function Node	AI Agent Node
What executes	JavaScript in sandbox	Full LLM + tool call loop
Determinism	Fully deterministic	Non-deterministic (LLM output varies)
Cost	Zero LLM API cost	LLM API cost per call
Latency	Milliseconds	Seconds (LLM + tools)
Best for	Calculations, transformations, validation	Natural language processing, decision-making, multi-step tasks

Error Handling in Workflow

If an AI Function returns an error object or throws a runtime exception, the node sets its status to Faulted. Workflow designers can add error branches from the AI Function node to handle failure gracefully:

// Function returns structured error:
return { error: "invalid_input", message: "startDate must be before endDate" };

// Node detects error key → sets node to Faulted state
// Workflow designer routes the "Error" branch to a notification step

← As MCP Tools Next: Security →