Starting Conversations with AI Assistants

After a successful anonymousLogin, you can initiate calls to your AI Assistant using the standard newCall method. The session is locked to the AI Assistant, so the destination parameter is ignored.

try client.newCall(
    callerName: String, // Display name (passed to AI assistant)
    callerNumber: String, // Caller number (passed to AI assistant)
    destinationNumber: String, // Ignored after anonymous login. All calls will be routed to the AI assistant
    callId: UUID, // Unique identifier for the call
    clientState: String? = nil, // Optional custom state information for your application
    customHeaders: [String: String] = [:] // Optional SIP headers to pass context to the AI assistant in the form of dynamic variables
)

Parameter	Type	Description
callerName	String	Your display name (passed to AI assistant)
callerNumber	String	Your phone number (passed to AI assistant)
destinationNumber	String	Ignored after anonymous login - can be empty string
callId	UUID	Unique identifier for the call
clientState	String?	Optional custom state information for your application
customHeaders	[String: String]	Optional SIP headers to pass context to the AI assistant (mapped to dynamic variables)

Note that you can also provide customHeaders in the newCall method. These headers need to start with the X- prefix and will be mapped to dynamic variables in the AI assistant (e.g., X-Account-Number becomes {{account_number}}). Hyphens in header names are converted to underscores in variable names.

// After a successful anonymousLogin...

do {
    let call = try client.newCall(
        callerName: "John Doe",
        callerNumber: "+1234567890",
        destinationNumber: "", // Destination is ignored, can be empty
        callId: UUID(),
        clientState: "ai_conversation_session",
        customHeaders: [
            "X-Session-Context": "support_request",
            "X-User-Tier": "premium"
        ]
    )
} catch {
    print("Failed to start call: \(error)")
}

class AIAssistantManager {
    private let client = TxClient()
    private var currentCall: Call?

    init() {
        client.delegate = self
    }

    func startAIConversation(assistantId: String) {
        // Step 1: Anonymous login
        client.anonymousLogin(targetId: assistantId)
    }

    private func startCall() {
        do {
            currentCall = try client.newCall(
                callerName: "Customer",
                callerNumber: "+1234567890",
                destinationNumber: "", // Ignored
                callId: UUID(),
                clientState: "ai_session",
                customHeaders: [
                    "X-Session-Context": "customer_support",
                    "X-User-Tier": "premium"
                ]
            )
        } catch {
            print("Failed to start call: \(error)")
        }
    }
}

extension AIAssistantManager: TxClientDelegate {
    func onClientReady() {
        // Login successful, start the call
        startCall()
    }

    func onCallStateUpdated(callState: CallState, callId: UUID) {
        guard currentCall?.callId == callId else { return }

        switch callState {
        case .NEW:
            print("Call invitation sent")
        case .RINGING:
            print("AI Assistant ringing...")
        case .ACTIVE:
            print("Connected to AI Assistant")
            // Start listening for transcripts
        case .DONE:
            print("AI conversation ended")
        default:
            break
        }
    }
}

• Automatic Answer: AI assistants automatically answer calls - no manual answer required
• Destination Ignored: The destinationNumber parameter is ignored after anonymous login
• Call Routing: All calls are routed to the AI assistant specified during login
• Standard Controls: Use existing call management methods (mute, hold, end call)
• Custom Headers: You can pass custom SIP headers to provide context to the AI assistant. They will be mapped to dynamic variables in the portal. Hyphens in header names are converted to underscores in variable names, e.g. X-Session-Context header maps to {{session_context}} variable.

Monitor call states as you would with regular calls:

func onCallStateUpdated(callState: CallState, callId: UUID) {
    guard currentCall?.callId == callId else { return }

    switch callState {
    case .NEW:
        print("Calling AI Assistant...")
    case .RINGING:
        print("AI Assistant ringing...")
    case .ACTIVE:
        print("Connected to AI Assistant")
        // Start listening for transcripts
        setupTranscriptListener()
    case .DONE:
        print("AI conversation ended")
    default:
        break
    }
}

Handle call-related errors:

func onClientError(error: Error) {
    if let txError = error as? TxError {
        switch txError {
        case .callFailure:
            print("Failed to start conversation with AI Assistant")
        default:
            print("Call error: \(txError)")
        }
    }
}

Once connected, use standard call management methods:

// Get the active call
guard let activeCall = currentCall else { return }

// Mute/unmute
activeCall.muteUnmute()

// Hold/unhold
activeCall.holdUnhold()

// End call
activeCall.hangup()

After starting a conversation:

1. Set up transcript updates to receive real-time conversation data
2. Send text messages during the active call
3. Use standard call controls for mute, hold, and end call operations