AI Outbound Agent State

The AIOutboundAgentState extends the regular AI Agent state to automate outbound interactions—e.g., phone calls, chat messages, or messaging-app conversations—directly from a workflow. In addition to the usual LLM configuration, tools, and outcomes, the state lets you specify:

• Outbound channel details (phone, Zalo, WhatsApp, Telegram, …) via outboundConfig

• Realtime voice features (STT/TTS/VAD) via voiceConfig

AIOutboundAgentState

Parameter	Description	Type	Required
agentName	The name of the agent.	string	yes
aiModel	The name of AI Language Model. Default value is 'gpt-4o'.	string	no
llmConfig	The configuration for the language model.	object	no
systemMessage	The system message used for constructing LLM prompt. Defaults to "You are a helpful AI Assistant."	string	yes
userMessage	The user message.	string	yes
maxToolExecutions	The maximum number of tool executions. Default is 10.	integer	no
memory	The memory of the agent. If not specify, the workflow process instance scope is used.	object	no
output	JSON schema for agent data output. See AgentDataOutput.	object	yes
tools	Define list of tools. Each tool is described by the ToolForAI schema.	array	no
agentOutcomes	Define list of agent outcomes. Each outcome is described by the OnAgentOutcome schema.	array	yes
dataFilter	Filter to apply to the state data.	string	no
outboundConfig	Channel-specific outbound settings.	object	yes
voiceConfig	Voice features (STT, TTS, VAD) for realtime calls.	object	no

LLMConfig

The same as LLMConfig from AIAgent State

ChatMemory

The same as ChatMemory from AIAgent State

AgentDataOutput

The same as AgentDataOutput from AIAgent State

OnAgentOutcome

The same as OnAgentOutcome from AIAgent State

ToolForAI

The same as ToolForAI from AIAgent State

OutboundConfig

The OutboundConfig defines the channel-specific outbound settings.

Property	Type	Description	Required
greeting	string	The static greeting message to be used by the agent. This message is used when the agent is first initialized.	no
greetingInstructions	string	The instructions for the LLM to use when generating the greeting message. This configuration takes precedence over the static greeting message.	no
outboundTarget	object	The outbound target configuration. This configuration is used to define the target for the outbound agent.	yes

OutboundTarget

The OutboundTarget defines the target for the outbound agent.

Property	Type	Description	Required
targetType	string	The type of the outbound target. This can be voice, zalo, whatsapp, etc. Default is voice.	yes
targetAddress	string	The address of the outbound target. This can be an email address, phone number, Zalo ID, etc. The format depends on the target type.	yes
targetName	string	The name of the outbound target. This is used for identification purposes.	yes

VoiceConfig

VoiceConfig is the single block that tells the workflow how to listen, think, and speak during a telephone or voice-chat session. Because speech is a round-trip of audio → text → LLM → text → audio, VoiceConfig is split into four conceptual sub-modules, each matching one step in that loop:

1. VAD – “Is anyone talking right now?”

2. STT – “What did they just say?”

3. LLM – “How should the agent respond?”

4. TTS – “Say it out loud—in a human voice.”

The pipeline looks like this:

Caller audio → VAD → STT ─┐
                         ├─► LLM (reasoning/JSON tools)
LLM reply ◄──────────────┘
LLM reply → TTS → Agent audio

We can mix-and-match providers for every step; each has its own latency, cost, language coverage, and feature set.

Why each component matters

Voice-Activity Detection (VAD)

Purpose: Detects the precise start and end of human speech in the inbound audio stream. Why it’s critical: If VAD fires too late you waste the caller’s first syllables; if it fires too early you feed silence or background noise into STT and spend tokens on “uh …”. Good VAD also enables barge-in (interrupting TTS mid-sentence) and double-talk detection.

Typical knobs inside the vad block

• provider name (silero is the default implementation)

• energy / probability thresholds

• timeouts for “no-speech” and “end-of-speech”

Speech-to-Text (STT)

Purpose: Transforms raw audio chunks into partial and final transcripts. Why it’s critical: Whatever the LLM “hears” comes from STT; recognition accuracy drives the entire conversational quality. Latency drives perceived responsiveness.

Key configuration areas

• Provider & model – e.g. OpenAI Whisper large-v3, Deepgram Nova-2, Google STT tel-alpha

• Language/locale – supply a BCP-47 code like vi-VN or en-US so the model loads the right phoneme set

• Streaming vs batch – most providers stream; some cheaper models require a full clip upload

• Vocabulary bias / hints – business terms, proper names, SKU codes

• Post-processing – capitalization, profanity masking, punctuation injection

• Security – API key, private endpoint, or on-prem GPU deployment

Large-Language Model (LLM)

Purpose: Understands user intent, decides on tool calls, chooses the next action, and produces a textual reply (or a JSON payload if your state’s output schema demands structured data).

Inside a voice agent the LLM sits in the tightest latency loop after STT, so choosing how the LLM delivers its tokens changes the entire user experience:

• Realtime LLM

  • A realtime model ingests raw audio, reasons over it, and streams synthetic speech back without any external STT or TTS step.

  • What changes in the pipeline

    • No separate STT/TTS blocks. The model hears tone, hesitations, laughter—cues that are normally lost in a transcript.

    • Built-in turn detection. Most providers decide when you’ve finished speaking; We recommends relying on that internal detector. If you want to fall back to default turn-detector you must still bolt on an STT plugin so the detector can read interim transcripts.

    • No hard-scripted speech. You can cue the model with instructions, but you cannot guarantee it will read a line verbatim. For legally approved disclaimers, attach a conventional TTS plugin and use greetingInstructions for that segment.

• Non-realtime LLM (Classic)

The classic voice-AI stack separates concerns:

Caller audio → VAD → STT → text
                          ↓
                       LLM  ← current context & tools
                          ↓
Reply text → TTS → audio to caller

Why it’s still popular:

Advantage	Implication
Deterministic text flow. Every turn yields clean, timestamped transcripts.	Great for analytics, compliance, post-call RAG pipelines.
Fine-grained control. You choose best-of-breed STT, specialised LLM tooling, and premium or budget TTS per use-case.	Extra integration work and ~1-2 s additional latency.
Script fidelity. A TTS engine will read a legal disclaimer exactly as written.	Voices may sound less expressive unless you invest in neural styles.

Text-to-Speech (TTS)

Purpose: Transforms the LLM’s textual reply into audio the caller hears. Why it’s critical: Humans judge “bot-ness” mainly by voice quality and timing. A 220 ms chunk-synthesis delay feels natural; 800 ms feels robotic.

TTS options worth documenting

• Voice/character – Rachel, en-US-Wavenet-D, Alloy-en-v2

• Style & prosody controls – speaking rate, pitch, emotion, stability, pronunciation lexicons

• Streaming support – mandatory for realtime pipelines; optional for batch

• Silence trimming & filler – some providers auto-trim leading breaths; some insert breathing/fillers you may want to disable

• Bandwidth – telephony lines are 8 kHz mono; web or app can handle 22 kHz stereo

VoiceConfig Properties

Property	Type	Description	Required
stt	object	The speech-to-text configuration (Optional).	no
tts	object	The text-to-speech configuration (Optional).	no
vad	object	The voice activity detection configuration (Optional).	no
allowInterruptions	boolean	Whether to allow interruptions during the voice interaction. Default is false.	no

STT

The STT defines the configuration for the Speech To Text (STT) to be used by the AI Agent.

Property	Type	Description	Required
provider	string	The name of the STT provider. Allowed values: 'deepgram', 'openai', 'google', 'elevenlabs', 'fal', 'groq'. This determines which STT backend will be used.	yes
model	string	The model to use for speech recognition. This is provider-specific.	no
language	string	The language code for recognition. This is provider-specific. Example: 'en-US', 'vi-VN', etc.	no
apiKey	string	The API key or credentials for the STT service. This is required for most providers to authenticate requests.	no
baseUrl	string	The base URL for the STT service. This is used for custom endpoints or self-hosted deployments. Optional for most cloud providers.
providerOptions	object	Provider-specific configuration options for STT. Use this to supply additional settings required by your provider.

Supported options by STT provider:

• deepgram:

detect_language: Whether to enable automatic language detection. Defaults to false.
interim_results: Whether to return interim (non-final) transcription results. Defaults to true.
punctuate: >-
  Whether to add punctuations to the transcription. Defaults to true. Turn
  detector will work better with punctuations.
smart_format: Whether to apply smart formatting to numbers, dates, etc. Defaults to true.
sample_rate: The sample rate of the audio in Hz. Defaults to 16000.
endpointing_ms: >-
  Time in milliseconds of silence to consider end of speech. Set to 0 to
  disable. Defaults to 25.
filler_words: >-
  Whether to include filler words (um, uh, etc.) in transcription. Defaults to
  true.
profanity_filter: Whether to filter profanity from the transcription. Defaults to false.
numerals: Whether to include numerals in the transcription. Defaults to false.
mip_opt_out: Whether to take part in the model improvement program, Defaults to false.

Example:

"stt": {
    "provider": "groq",
    "model": "whisper-large-v3-turbo",
    "language": "vi",
    "apiKey": "${ $SECRETS.GROQ_API_KEY }"
}

• openai:

detect_language: Whether to enable automatic language detection. Defaults to false.

• google:

languages: List of language codes to recognize, Google STT can accept multiple languages
detect_language: Whether to enable automatic language detection. Defaults to false.
credentials_info: >-
  Google credentials info. This is a JSON string with the following fields:
  project_id, client_email, private_key_id, private_key. See
  https://cloud.google.com/docs/authentication/getting-started for more
  information.
credentials_file: URL to a file containing Google credentials.
credentials_file_auth_headers: HTTP headers to use when downloading the credentials file.

• elevenlabs: None

• fal: None

• groq: None

Example:

"stt": {
    "provider": "groq",
    "model": "whisper-large-v3-turbo",
    "language": "vi",
    "apiKey": "${ $SECRETS.GROQ_API_KEY }"
}

• azure: None

"stt": {
    "provider": "azure",
    "providerOptions": {
        "speech_key": "${ $SECRETS.AZURE_SPEECH_KEY }",
        "speech_region": "${ $SECRETS.AZURE_SPEECH_REGION }"
    }
}

TTS

The TTS defines the configuration for Text To Speech (TTS) to be used by the AI Agent

Property	Type	Description	Required
provider	string	The name of the TTS provider. Allowed values: 'openai', 'deepgram', 'google', 'elevenlabs', 'groq'. This determines which TTS backend will be used.	no
model	string	The model to use for TTS. This is provider-specific.	no
voice	string	The voice to use for TTS. This is provider-specific and may refer to a named voice (e.g., 'en-US-Wavenet-D' for Google, 'Rachel' for ElevenLabs, etc.).	no
language	string	The language code for TTS. This is provider-specific and may affect pronunciation and available voices.	no
apiKey	string	The API key or credentials for the TTS service. This is required for most providers to authenticate requests.	no
baseUrl	string	The base URL for the TTS service. This is used for custom endpoints or self-hosted deployments. Optional for most cloud providers.	no
providerOptions	object	Provider-specific configuration options for TTS. Use this to supply additional settings required by your provider.	no

Supported options by TTS provider:

• openai:

speed: Speaking speed
instructions: >-
  Instructions to control tone, style, and other characteristics of the speech.
  Does not work with tts-1 or tts-1-hd models

• deepgram:

encoding: Audio encoding, eg: linear16
sample_rate: Sample rate, eg: 24000

• google:

gender: Voice gender. Valid values are male, female, and neutral.
credentials_info: >-
  Google credentials info. This is a JSON string with the following fields:
  project_id, client_email, private_key_id, private_key. See
  https://cloud.google.com/docs/authentication/getting-started for more
  information.
credentials_file: URL to a file containing Google credentials.
credentials_file_auth_headers: HTTP headers to use when downloading the credentials file.

Example:

"tts": {
    "provider": "google",
    "voice": "vi-VN-Standard-A",
    "apiKey": "https://xbot-uat.hcm03.vstorage.vngcloud.vn/xbot-out-livekit-uat.json?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Date=20250516T102347Z&X-Amz-SignedHeaders=host&X-Amz-Expires=518699&X-Amz-Credential=e45eff37790ca1f6a4e0174698dc9991%2F20250516%2FHCM03%2Fs3%2Faws4_request&X-Amz-Signature=84b2d4a465379427eeac86c0210d954ec6a6e7ac2e869600377134311093ae89",
    "instructions": "Speak in a friendly and professional manner. Use Vietnamese language to communicate with the customer."
}

• elevenlabs:

voice_settings: >-
  Voice settings object: stability, similarity_boost, style, use_speaker_boost,
  speed. See
  https://elevenlabs.io/docs/api-reference/text-to-speech/convert#request.body.voice_settings
streaming_latency: >-
  Optimize for streaming latency, defaults to 0 - disabled. 4 for max latency
  optimizations

• groq: None

• azure:

Example:

"tts": {
    "provider": "azure",
    "providerOptions": {
        "speech_key": "${ $SECRETS.AZURE_SPEECH_KEY }",
        "speech_region": "${ $SECRETS.AZURE_SPEECH_REGION }"
    }
}

VAD

The VAD defines the configuration for the Voice Activity Detection (VAD) to be used by the AI Agent

Property	Type	Description	Required
provider	string	The name of the provider. Default is 'silero'.	no

Example:

YAML

code: BANKING_AGENT name: Banking aGENT agents: - type: aiagent name: AgentSelector start: true transition: end: false targetId: RagAgent answerAfterFinish: false answerMessage: null aiModel: gpt-4o-mini llmConfig: provider: openai apiKey: '${ $SECRETS.OPENAI_API_KEY }' overrideParams: model: gpt-4o-mini temperature: 0 topP: 0 'n': 0 logprobs: 0 echo: false stop: [] maxTokens: 1024 presencePenalty: 0 frequencyPenalty: 0 logitBias: null required: null systemMessage: >- You are the VietTrust Bank assistant, helping to address queries related to information regarding the bank’s offerings, including branch locations, account types, savings products, loan services, credit card options, and other banking solutions. Select the most appropriate tool based on its description, and return the name of the selected tool. Bellow is tools list: - RagAgent: This tool specializes in answering queries regarding the bank’s offerings, including branch locations, account types, savings products, loan services, credit card options, and other banking solutions. Use this tool to access up-to-date data from the system's knowledge base. - OutboundCollectorAgent: This tool is designed to assist with outbound collection tasks, such as sending reminders or notifications to customers regarding their account status or payment due dates. Use this tool for tasks related to customer outreach and communication. userMessage: '${.request.question}' maxToolExecutions: 10 memory: memoryId: '${ $conversationId + "-AgentSelector" }' memoryType: message_window maxMessages: 5 maxTokens: null memoryOptimizer: null output: schema: |- { "type": "object", "properties": { "selectedAgent": { "type": "string", "description": "The exact name of agent to be selected." } }, "required": [ "selectedAgent" ] } tools: [] askUserForToolsInput: false agentOutcomes: - condition: >- ${ ( ($agentOutcome.returnValues.selectedAgent != null) and (($agentOutcome.returnValues.selectedAgent | contains("OutboundCollectorAgent")) == true) ) or ( ($agentOutcome.returnValues.properties.selectedAgent != null) and (($agentOutcome.returnValues.properties.selectedAgent | contains("OutboundCollectorAgent")) == true) ) } finish: true transition: end: false targetId: OutboundCollectorAgent answerAfterFinish: false answerMessage: null - condition: >- ${ ( ($agentOutcome.returnValues.selectedAgent != null) and (($agentOutcome.returnValues.selectedAgent | contains("RagAgent")) == true) ) or ( ($agentOutcome.returnValues.properties.selectedAgent != null) and (($agentOutcome.returnValues.properties.selectedAgent | contains("RagAgent")) == true) ) } finish: true transition: end: false targetId: RagAgent answerAfterFinish: false answerMessage: null - condition: '${true}' finish: true transition: end: false targetId: RagAgent answerAfterFinish: false answerMessage: null loginRequired: null - type: kbagent name: RagAgent start: false transition: end: true targetId: null answerAfterFinish: true answerMessage: '${ $RagAgent.outputMessage }' systemMessage: >- You are an AI assistant for **VietTrust Bank**, a modern bank that offers a wide range of financial services. Your role is to assist customers by providing accurate and helpful information regarding the bank’s offerings, including branch locations, account types, savings products, loan services, credit card options, and other banking solutions. ### Instructions: 1. **Branch Locations:** - Provide customers with details about the nearest branch based on the list of locations in the context. Use the city, district, or address details provided by the customer to find the nearest branch. 2. **Account Types:** - Explain the available types of bank accounts, including **savings accounts**, **current accounts**, and **foreign currency accounts**. Provide details such as minimum balance requirements, interest rates, and benefits for each type of account. 3. **Savings Products:** - Offer information about **savings products**, including fixed-term deposits and special savings plans. Provide details such as interest rates, account terms, and any specific conditions (e.g., minimum deposit amounts). 4. **Loan Products:** - Assist customers in understanding the bank’s **loan offerings**, including **personal loans**, **home loans**, **auto loans**, and **business loans**. Mention the interest rates, repayment periods, and any special conditions based on the loan type. 5. **Credit Cards:** - Provide information about the bank’s **credit card options**, including **credit limits**, **annual fees**, and **reward programs**. Highlight specific benefits for different types of cards, such as Visa, MasterCard, or co-branded cards. 6. **Banking Services:** - Explain additional banking services, such as **online banking**, **mobile banking**, **money transfers**, and **bill payment** services. - Guide customers through the process of setting up or managing these services if the context provides instructions. 7. **Priority Banking:** - For **priority customers**, provide details about exclusive services such as higher interest rates, lower service fees, and access to dedicated account managers. If the customer qualifies, inform them about the advantages of priority banking. 8. **Cross-References:** - If a question requires information from multiple sections, provide cross-references to relevant sections (e.g., linking loan products with interest rates or providing details about applicable savings account terms). 9. **Missing Context:** - If a customer asks for information that is not available in the provided context, respond politely with: **"Xin lỗi, tôi không có thông tin để trả lời câu hỏi này."** ("Sorry, I don’t have the information to answer this question."). 10. **Use Only Provided Context:** - You must **only use the information provided in the context** to answer the questions. Do not use any external knowledge or assumptions. If the required information is not available in the context, respond politely with, **"Xin lỗi, tôi không có thông tin để trả lời câu hỏi này."** ("Sorry, I don’t have the information to answer this question."). ### Example Queries: - "Where is the nearest VietTrust Bank branch in Hanoi?" - "What is the interest rate for a 12-month savings account?" - "Can I apply for a home loan online?" - "What credit card options do you offer with no annual fee?" - "How do I transfer money to another bank account?" ### Language and Tone: - Provide all responses in **Vietnamese**, ensuring that they are clear, polite, and strictly based on the context provided by VietTrust Bank. If the required information is missing, respond politely and do not make any assumptions. userMessage: '${.request.question}' llmConfig: provider: groq apiKey: '${ $SECRETS.GROQ_API_KEY }' overrideParams: model: meta-llama/llama-4-scout-17b-16e-instruct temperature: 0 topP: 0 'n': 0 logprobs: 0 echo: false stop: [] maxTokens: 512 presencePenalty: 0 frequencyPenalty: 0 logitBias: null required: null memory: memoryId: '${ $conversationId }' memoryType: message_window maxMessages: 5 maxTokens: null knowledgeBase: queryStrategy: null knowledgeBaseCodes: - XFILE_BANKING ragConfig: history: messageLimit: 5 retriever: maxResults: 5 minScore: 0.6 kbLocalRerank: true includeDocReference: true ragMode: NAIVE loginRequired: null streaming: false - type: aiagent name: OutboundCollectorAgent start: false transition: end: false targetId: OutboundAgent answerAfterFinish: false aiModel: gpt-4.0-mini llmConfig: provider: openai apiKey: '${ $SECRETS.OPENAI_API_KEY }' overrideParams: model: gpt-4o-mini temperature: 0 topP: 0 'n': 0 logprobs: 0 echo: false stop: [] maxTokens: 1268 presencePenalty: 0 frequencyPenalty: 0 logitBias: null required: null systemMessage: >- You are AI Assistant, dedicated to collect the information. Your task is to collect the outbound call information based on the question from the user. You will be provided with the outbound call information, and you need to ask the user for any missing information. userMessage: '${.request.question}' maxToolExecutions: 40 memory: memoryId: '${ $conversationId }' memoryType: message_window maxMessages: 19 maxTokens: null output: schema: >- {"type":"object","properties":{"customerName":{"type":"string"},"phoneNumber":{"type":"string"},"message":{"type":"string"}},"required":["customerName","phoneNumber","message"]} tools: [] askUserForToolsInput: true agentOutcomes: [] loginRequired: null - type: aioutboundagent name: OutboundAgent start: false transition: end: true targetId: null systemMessage: >- You are a conversational AI agent for **VietTrust Bank**, a modern financial institution offering a full suite of banking services. Your job is to place outbound calls to customers in order to deliver polite, professional reminders or notifications. For each call you will be given: - Customer’s profile (name, account/loan details, etc.) - The scripted message you need to convey - The customer’s phone number Always speak in Vietnamese, using a courteous and friendly tone appropriate for a bank representative. Before ending each call, always thank the customer for their time. Your task is informing the customer the message from following instructions: {{.request.question}} == Message: == {{ $OutboundCollectorAgent.returnValues.message }}. You can also use provided tools to answer the customer's queries (if any) based on known information and don't ask it again: - Customer phone: ${ $OutboundCollectorAgent.returnValues.phoneNumber }. maxToolExecutions: 10 llmConfig: provider: google apiKey: '${ $SECRETS.GEMINI_API_KEY }' overrideParams: model: gemini-2.5-flash-preview-04-17 temperature: 0 topP: 0 'n': 0 logprobs: 0 echo: false stop: [] maxTokens: 512 presencePenalty: 0 frequencyPenalty: 0 logitBias: null required: null memory: memoryId: '${ $conversationId }' memoryType: message_window maxMessages: 5 maxTokens: null tools: - name: getCustomerInfoByPhone description: Get customer information based on phone number. parameters: schema: |- { "type": "object", "properties": { "phone": {"type": "string", "description": "Customer phone number."} }, "required": ["phone"] } output: schema: |- { "type": "object", "properties": { "customerCode": {"type": "string"}, "fullName": {"type": "string"}, "phoneNumber": {"type": "string"}, "address": {"type": "string"} }, "required": ["customerCode","fullName","phoneNumber","address"] } execution: actionMode: sequential actions: - id: null name: getCustomerInfoByPhone condition: '${true}' functionRef: code: getCustomerInfo_bankingApi name: getCustomerInfo description: null asyncInvoke: false type: rest definition: type: simple-rest url: >- https://xplatform-api-uat.a4b.vn/partner/xfai/api/mock/banking/getCustomerInfo method: GET headers: xp-api-key: >- 0123456789 queryParams: phone: '${ .phone }' body: null auth: null inputs: - name: phone value: null arguments: phone: '${ $toolArguments.phone }' metadata: null - name: getAccountsByPhone description: Get all accounts by customer phone number. parameters: schema: |- { "type": "object", "properties": { "phone": {"type": "string"} }, "required": ["phone"] } output: schema: |- { "type": "array", "items": { "type": "object", "properties": { "accountId": {"type": "string"}, "customerCode": {"type": "string"}, "balance": {"type": "number"} }, "required": ["accountId", "customerCode", "balance"] } } execution: actionMode: sequential actions: - id: null name: getAccountsByPhone condition: '${true}' functionRef: code: getAccountsByPhone_bankingApi name: getAccountsByPhone description: null asyncInvoke: false type: rest definition: type: simple-rest url: >- https://xplatform-api-uat.a4b.vn/partner/xfai/api/mock/banking/getAccountsByPhone method: GET headers: xp-api-key: >- 0123456789 queryParams: phone: '${ .phone }' body: null auth: null inputs: - name: phone value: null arguments: phone: '${ $toolArguments.phone }' metadata: null - name: getLoansByCustomer description: Get all loans by customer code. parameters: schema: |- { "type": "object", "properties": { "customerCode": {"type": "string"} }, "required": ["customerCode"] } output: schema: |- { "type": "array", "items": { "type": "object", "properties": { "loanId": {"type": "string"}, "customerCode": {"type": "string"}, "principal": {"type": "number"}, "interestRate": {"type": "number"}, "termMonths": {"type": "integer"}, "outstanding": {"type": "number"}, "createdDate": {"type": "string"} }, "required": ["loanId", "customerCode", "principal", "interestRate", "termMonths", "outstanding", "createdDate"] } } execution: actionMode: sequential actions: - id: null name: getLoansByCustomer condition: '${true}' functionRef: code: getLoansByCustomer_bankingApi name: getLoansByCustomer description: null asyncInvoke: false type: rest definition: type: simple-rest url: >- https://xplatform-api-uat.a4b.vn/partner/xfai/api/mock/banking/getLoansByCustomer method: GET headers: xp-api-key: >- 0123456789 queryParams: customerCode: '${ .customerCode }' body: null auth: null inputs: - name: customerCode value: null arguments: customerCode: '${ $toolArguments.customerCode }' metadata: null - name: getLoanPayments description: Get all payments for a loan. parameters: schema: |- { "type": "object", "properties": { "loanId": {"type": "string"} }, "required": ["loanId"] } output: schema: |- { "type": "array", "items": { "type": "object", "properties": { "paymentId": {"type": "string"}, "date": {"type": "string"}, "amount": {"type": "number"} }, "required": ["paymentId", "date", "amount"] } } execution: actionMode: sequential actions: - id: null name: getLoanPayments condition: '${true}' functionRef: code: getLoanPayments_bankingApi name: getLoanPayments description: null asyncInvoke: false type: rest definition: type: simple-rest url: >- https://xplatform-api-uat.a4b.vn/partner/xfai/api/mock/banking/getLoanPayments method: GET headers: xp-api-key: >- 0123456789 queryParams: loanId: '${ .loanId }' body: null auth: null inputs: - name: loanId value: null arguments: loanId: '${ $toolArguments.loanId }' metadata: null output: schema: >- {"type":"object","properties":{"answer":{"type":"string"}},"required":["answer"]} streaming: false voiceConfig: allowInterruptions: false tts: provider: google voice: vi-VN-Standard-A apiKey: >- https://xbot-uat.hcm03.vstorage.vngcloud.vn/xbot-out-livekit-uat.json?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Date=20250516T102347Z&X-Amz-SignedHeaders=host&X-Amz-Expires=518699&X-Amz-Credential=e45eff37790ca1f6a4e0174698dc9991%2F20250516%2FHCM03%2Fs3%2Faws4_request&X-Amz-Signature=84b2d4a465379427eeac86c0210d954ec6a6e7ac2e869600377134311093ae89 instructions: >- Speak in a friendly and professional manner. Use Vietnamese language to communicate with the customer. stt: provider: groq model: whisper-large-v3-turbo language: vi apiKey: '${ $SECRETS.GROQ_API_KEY }' outboundConfig: greetingInstructions: >- Greet the customer then introduce yourself as a representative of VietTrust Bank. Ask if they have time to talk. Use Vietnamese language to communicate with the customer. outboundTarget: targetType: voice targetAddress: '${ $OutboundCollectorAgent.returnValues.phoneNumber }' targetName: '${ $OutboundCollectorAgent.returnValues.customerName }' appCode: XCHATBOT tenantCode: DEMO

---

This document provides a detailed view of the AIOutboundAgentState state and its related objects, including comprehensive schema definitions, required fields, and descriptions for each attribute within the AIOutboundAgentState and associated schemas. This specification ensures clarity and completeness for integrating Outboudn AI agents within serverless workflows.