Prefab Agents Guide

Complete guide to the pre-built Prefab agents included in the SignalWire AI Agents TypeScript SDK.

export {}; // treat each example as a module (top-level await)
declare global {
  const InfoGathererAgent: typeof import('@signalwire/sdk').InfoGathererAgent;
  const SurveyAgent: typeof import('@signalwire/sdk').SurveyAgent;
  const FAQBotAgent: typeof import('@signalwire/sdk').FAQBotAgent;
  const ConciergeAgent: typeof import('@signalwire/sdk').ConciergeAgent;
  const ReceptionistAgent: typeof import('@signalwire/sdk').ReceptionistAgent;
}

Overview
InfoGathererAgent
SurveyAgent
FAQBotAgent
ConciergeAgent
ReceptionistAgent
Factory Functions
Subclassing Prefabs

Overview

Prefab agents are ready-to-use agent implementations built on top of AgentBase. They provide complete, production-ready AI voice agents for common use cases -- information gathering, surveys, FAQ bots, department routing, and front-desk reception.

Each prefab:

Extends AgentBase and inherits all core functionality (HTTP serving, SWML generation, basic auth, CORS, proxy detection).
Defines its own SWAIG tools through the defineTools() override, giving the AI model the ability to perform domain-specific actions.
Declares static PROMPT_SECTIONS that set up the AI's role and behavioral rules automatically.
Adds dynamic prompt sections in the constructor based on your configuration (field lists, question lists, department directories, etc.).
Tracks per-call session state using call_id from the incoming request data, so multiple concurrent calls are isolated.

Prefabs can be used in two ways:

Directly via the constructor -- pass a configuration object and get a fully functional agent.
Via a factory function -- a shorthand that creates and returns a new instance (e.g., createInfoGathererAgent(config)).

All prefabs are exported from the main SDK entry point:

import {
  InfoGathererAgent, SurveyAgent, FAQBotAgent,
  ConciergeAgent, ReceptionistAgent,
} from '@signalwire/sdk';

InfoGathererAgent

A conversational agent that asks the caller a sequence of questions one at a time and records each answer in global_data. Supports both static configuration (questions defined at construction) and dynamic configuration (questions resolved per request via a callback).

Source: src/prefabs/InfoGathererAgent.ts

InfoGatherer Configuration

The constructor accepts an InfoGathererConfig object:

Property	Type	Required	Default	Description
`name`	`string`	No	`"info_gatherer"`	Agent display name.
`route`	`string`	No	`"/info_gatherer"`	HTTP route for this agent.
`questions`	`InfoGathererQuestion[]`	No	--	Questions to ask (static mode). Omit for dynamic mode.
`questionCallback`	`InfoGathererQuestionCallback`	No	--	Resolves questions per request (dynamic mode); equivalent to calling `setQuestionCallback()` after construction.
`agentOptions`	`Partial<AgentOptions>`	No	--	Additional AgentBase options forwarded to `super()` (e.g., `port`, `basicAuth`).

InfoGatherer Questions

Each entry in the questions array is an InfoGathererQuestion:

Property	Type	Required	Default	Description
`key_name`	`string`	Yes	--	Identifier used as the key when storing the caller's answer.
`question_text`	`string`	Yes	--	The question text spoken to the caller.
`confirm`	`boolean`	No	`false`	When `true`, the agent insists the caller confirms the answer before submitting.

InfoGatherer Tools

The agent registers two SWAIG tools:

`start_questions`

Retrieves the first question from global_data.questions and returns an instruction asking the caller to answer it.

Parameters: None.

Returns: A formatted instruction including the first question's text and confirmation guidance when confirm is set. Also sets replace_in_history to the generic "Welcome" prompt.

`submit_answer`

Records the caller's answer to the current question and advances to the next.

Parameters:

Name	Type	Required	Description
`answer`	`string`	Yes	The caller's answer to the current question.

Behavior:

Appends { key_name, answer } to global_data.answers.
Increments global_data.question_index.
If more questions remain, returns the next question's instruction text.
If all questions are answered, returns the completion message.

Dynamic Mode

Omit questions and register a callback to resolve the question list per incoming request:

const agent = new InfoGathererAgent({ name: 'dynamic-intake' });
agent.setQuestionCallback((queryParams, bodyParams, headers) => {
  if (queryParams['set'] === 'support') {
    return [
      { key_name: 'name', question_text: 'What is your name?' },
      { key_name: 'issue', question_text: "What's the issue?" },
    ];
  }
  return [{ key_name: 'name', question_text: 'What is your name?' }];
});

The callback runs on every SWML request, and its return value populates global_data.questions for that call. If no callback is registered or the callback throws, a default two-question fallback (name, message) is used.

InfoGatherer Example

import { InfoGathererAgent } from '@signalwire/sdk';

const agent = new InfoGathererAgent({
  name: 'PatientIntake',
  questions: [
    { key_name: 'full_name', question_text: 'What is your full legal name?' },
    { key_name: 'date_of_birth', question_text: 'What is your date of birth?', confirm: true },
    { key_name: 'phone_number', question_text: 'What is a good callback number?', confirm: true },
    { key_name: 'insurance_provider', question_text: 'Who is your insurance provider?' },
  ],
  agentOptions: {
    port: 3001,
    route: '/intake',
  },
});

agent.run();

SurveyAgent

A conversational agent that conducts surveys with multiple question types, branching logic based on answers, and per-answer scoring.

Source: src/prefabs/SurveyAgent.ts

Survey Configuration

The constructor accepts a SurveyConfig object:

Property	Type	Required	Default	Description
`name`	`string`	No	`"survey"`	Agent display name.
`route`	`string`	No	`"/survey"`	HTTP route for this agent.
`surveyName`	`string`	Yes	--	Human-readable survey name used in prompts and global data.
`questions`	`SurveyQuestion[]`	Yes	--	Ordered list of survey questions.
`introduction`	`string`	No	`"Welcome to our ${surveyName}. We appreciate your participation."`	Opening message before the first question (used as a non-bargeable static greeting).
`conclusion`	`string`	No	`"Thank you for completing our survey. Your feedback is valuable to us."`	Message spoken after the survey is complete.
`brandName`	`string`	No	`"Our Company"`	Brand or company name the agent represents.
`maxRetries`	`number`	No	`2`	Maximum number of times to retry invalid answers.
`onComplete`	`(responses: Record<string, unknown>, score: number) => void \| Promise<void>`	No	--	Callback fired when the survey is finished. Receives all responses and the total score.
`agentOptions`	`Partial<AgentOptions>`	No	--	Additional AgentBase options.

Each SurveyQuestion has the following shape:

Property	Type	Required	Description
`id`	`string`	Yes	Unique question identifier.
`text`	`string`	Yes	The question text to ask the caller.
`type`	`'multiple_choice' \| 'open_ended' \| 'rating' \| 'yes_no'`	Yes	Question type; determines validation and display.
`options`	`string[]`	No	Answer options (used when `type` is `multiple_choice`).
`scale`	`number`	No	For `rating` questions, the upper bound of the 1..scale range. Defaults to `5`.
`required`	`boolean`	No	Whether the question requires an answer. Defaults to `true`.
`nextQuestion`	`string \| Record<string, string>`	No	Controls flow after this question (see Branching Logic).
`points`	`number \| Record<string, number>`	No	Points awarded for answers (see Scoring).

Survey Question Types

Type	Validation	Notes
`multiple_choice`	Answer must exactly match one of the `options` (case-insensitive).	All options are read aloud.
`open_ended`	No validation; any answer is accepted.	Free-form response.
`rating`	Must be an integer between 1 and the question's `scale` (default `5`).	The AI specifies the scale to the caller.
`yes_no`	Must be a recognized affirmative or negative word.	Accepts: `yes`, `y`, `yeah`, `yep`, `sure`, `absolutely`, `correct`, `true`, `no`, `n`, `nah`, `nope`, `negative`, `false`. Normalized to `"yes"` or `"no"` before storage.

Survey Branching Logic

The nextQuestion property controls which question follows the current one:

Value	Behavior
`undefined` (omitted)	Proceed to the next question in array order. If this is the last question, the survey ends.
`string`	Always jump to the question with this ID, regardless of the answer.
`Record<string, string>`	Conditional branching: the answer value (case-insensitive) is looked up as a key, and the value is the next question ID. A special `_default` key serves as a fallback. If no key matches and there is no `_default`, falls through to the next question in array order.

Survey Scoring

The points property controls point accumulation:

Value	Behavior
`undefined` (omitted)	No points for this question.
`number`	Fixed points awarded for any answer.
`Record<string, number>`	Per-answer scoring: the answer value (case-insensitive) is looked up as a key, and the matched value is the number of points awarded. Unmatched answers receive 0 points.

The total score is accumulated across all answered questions and passed to the onComplete callback.

Survey Tools

The agent registers the following SWAIG tools:

`validate_response`

Validates whether a response satisfies the question's type and constraints without recording or advancing.

Parameters:

Name	Type	Required	Description
`question_id`	`string`	Yes	The ID of the question to validate against.
`response`	`string`	Yes	The candidate response.

Returns: A confirmation message when valid, or a type-specific error describing why the response is invalid.

`log_response`

Acknowledges that a validated response has been recorded for the specified question.

Parameters:

Name	Type	Required	Description
`question_id`	`string`	Yes	The ID of the question.
`response`	`string`	Yes	The validated response.

Returns: A confirmation message referencing the question's text.

`answer_question`

Records the caller's answer to the current survey question.

Parameters:

Name	Type	Required	Description
`question_id`	`string`	Yes	The ID of the question being answered.
`answer`	`string`	Yes	The caller's answer.

Behavior:

Validates the answer based on question type. Returns a validation error if invalid.
Normalizes the answer (e.g., yes/no variants become "yes" or "no").
Calculates and accumulates points.
Resolves the next question via branching logic.
If there is no next question, marks the survey complete, fires onComplete, and returns the completion message with answer count and total score.
Otherwise, advances the session and returns the next question text with its type-specific instructions.

`get_current_question`

Returns the current question to be asked to the caller.

Parameters: None.

Returns: The question ID, type, text, and type-specific instructions (options for multiple choice, scale for rating, etc.).

`get_survey_progress`

Returns the current progress of the survey.

Parameters: None.

Returns: A summary including questions answered out of total, percentage complete, current score, completion status, and all answers so far.

Survey Example

import { SurveyAgent } from '@signalwire/sdk';

const agent = new SurveyAgent({
  name: 'CustomerSatisfaction',
  surveyName: 'Customer Satisfaction Survey',
  introduction: 'Hi! We would love to hear your feedback about our service.',
  conclusion: 'Thanks for your feedback! We really appreciate it.',
  questions: [
    {
      id: 'overall',
      text: 'How would you rate your overall experience?',
      type: 'rating',
      points: { '9': 10, '10': 10, '7': 7, '8': 7 },
    },
    {
      id: 'recommend',
      text: 'Would you recommend us to a friend?',
      type: 'yes_no',
      points: { 'yes': 5, 'no': 0 },
      nextQuestion: {
        'yes': 'what_liked',
        'no': 'what_improve',
      },
    },
    {
      id: 'what_liked',
      text: 'What did you like the most about our service?',
      type: 'open_ended',
      nextQuestion: 'followup',
    },
    {
      id: 'what_improve',
      text: 'What could we improve?',
      type: 'open_ended',
      nextQuestion: 'followup',
    },
    {
      id: 'followup',
      text: 'How would you like us to follow up?',
      type: 'multiple_choice',
      options: ['Email', 'Phone call', 'No follow-up needed'],
    },
  ],
  onComplete: async (responses, score) => {
    console.log('Survey complete! Score:', score);
    console.log('Responses:', responses);
    // Store results, send to analytics, etc.
  },
});

agent.run();

FAQBotAgent

A conversational agent that answers frequently asked questions using keyword and word-overlap matching with configurable similarity thresholds and optional escalation to a live agent.

Source: src/prefabs/FAQBotAgent.ts

FAQBot Configuration

The constructor accepts a FAQBotConfig object:

Property	Type	Required	Default	Description
`name`	`string`	No	`"FAQBot"`	Agent display name.
`faqs`	`FAQEntry[]`	Yes	--	List of FAQ entries for the knowledge base.
`threshold`	`number`	No	`0.5`	Minimum match score (0--1) for an FAQ to be considered a match.
`escalationMessage`	`string`	No	`"I'm sorry, I couldn't find an answer to your question. Let me transfer you to someone who can help."`	Message spoken when no FAQ matches the query.
`escalationNumber`	`string`	No	--	Phone number to transfer to on escalation. If not set, the `escalate` tool is not registered.
`agentOptions`	`Partial<AgentOptions>`	No	--	Additional AgentBase options.

Each FAQEntry has the following shape:

Property	Type	Required	Description
`question`	`string`	Yes	The representative question text.
`answer`	`string`	Yes	The answer to provide when this FAQ matches.
`keywords`	`string[]`	No	Additional keywords to boost matching accuracy.

FAQBot Matching Engine

The matching engine computes a word-overlap similarity score between the caller's query and each FAQ entry.

Tokenization:

Text is lowercased.
Non-alphanumeric characters are replaced with spaces.
Text is split on whitespace.
Words shorter than 2 characters are removed.
Common English stop words are filtered out (e.g., "the", "is", "to", "and", "or", etc.).

Scoring:

Question overlap -- the number of shared tokens between the query and the FAQ question text, normalized by the smaller token set.
Keyword overlap -- each keyword is checked for substring presence in the lowercased query; the hit count is divided by total keywords.
Combined score -- if keywords are present, the final score is 0.6 * questionOverlap + 0.4 * keywordScore. Without keywords, only question overlap is used.

The best-scoring FAQ is returned if its score meets or exceeds the configured threshold. Runner-up matches above the threshold are mentioned as related results.

FAQBot Tools

`search_faq`

Searches the FAQ knowledge base for an answer to the caller's question.

Parameters:

Name	Type	Required	Description
`query`	`string`	Yes	The caller's question or search query.

Returns: The best matching FAQ entry with confidence score, plus up to 2 related runner-ups if they also exceed the threshold. If no match meets the threshold, returns the escalation message.

`escalate`

Transfers the caller to a live agent. Only registered when escalationNumber is configured.

Parameters:

Name	Type	Required	Description
`reason`	`string`	No	The reason for escalation. Defaults to `"Caller needs assistance beyond FAQ"`.

Behavior: Uses FunctionResult.connect() to transfer the call to the configured escalationNumber.

FAQBot Example

import { FAQBotAgent } from '@signalwire/sdk';

const agent = new FAQBotAgent({
  name: 'HelpDesk',
  threshold: 0.4,
  escalationNumber: '+15559876543',
  escalationMessage: 'I was not able to find an answer for that. Let me connect you with a support specialist.',
  faqs: [
    {
      question: 'What are your business hours?',
      answer: 'We are open Monday through Friday, 9 AM to 5 PM Eastern Time.',
      keywords: ['hours', 'open', 'close', 'schedule', 'when'],
    },
    {
      question: 'How do I reset my password?',
      answer: 'Go to the login page, click "Forgot Password", and follow the email instructions.',
      keywords: ['password', 'reset', 'forgot', 'login', 'locked'],
    },
    {
      question: 'What is your return policy?',
      answer: 'We offer a 30-day return policy for unused items in original packaging.',
      keywords: ['return', 'refund', 'exchange', 'policy'],
    },
    {
      question: 'How do I contact support?',
      answer: 'You can reach our support team by calling this number, by email at support@example.com, or through live chat on our website.',
      keywords: ['support', 'contact', 'help', 'email', 'chat'],
    },
  ],
  agentOptions: {
    route: '/helpdesk',
  },
});

agent.run();

ConciergeAgent

A virtual concierge for a venue or business. Provides information about services, amenities, and hours of operation, and answers availability and directions questions.

Source: src/prefabs/ConciergeAgent.ts

Concierge Configuration

The constructor accepts a ConciergeConfig object:

Property	Type	Required	Default	Description
`name`	`string`	No	`"concierge"`	Agent display name.
`route`	`string`	No	`"/concierge"`	HTTP route for this agent.
`venueName`	`string`	Yes	--	Name of the venue or business.
`services`	`string[]`	Yes	--	List of services offered.
`amenities`	`Record<string, Record<string, string>>`	Yes	--	Amenities as an object of amenity-name → detail-pairs.
`hoursOfOperation`	`Record<string, string>`	No	`{ default: '9 AM - 5 PM' }`	Operating hours by category.
`specialInstructions`	`string[]`	No	`[]`	Extra instruction bullets to append.
`welcomeMessage`	`string`	No	--	When set, installed as a non-bargeable static greeting.
`agentOptions`	`Partial<AgentOptions>`	No	--	Additional AgentBase options.

The venue name, all services, and amenity names are added as speech recognition hints.

Concierge Tools

`check_availability`

Checks whether a given service is offered on a specified date and time.

Parameters:

Name	Type	Required	Description
`service`	`string`	Yes	The service to check.
`date`	`string`	Yes	The date in `YYYY-MM-DD` format.
`time`	`string`	Yes	The time in `HH:MM` 24-hour format.

Returns: A confirmation or a list of offered services when the requested service is not available.

`get_directions`

Looks up directions for an amenity by name.

Parameters:

Name	Type	Required	Description
`location`	`string`	Yes	The amenity or location to get directions to.

Returns: Directions referencing the amenity's location detail, or a fallback pointing the caller to the front desk.

Concierge Example

import { ConciergeAgent } from '@signalwire/sdk';

const agent = new ConciergeAgent({
  venueName: 'Grand Hotel',
  services: ['room service', 'spa bookings', 'restaurant reservations'],
  amenities: {
    pool: { hours: '7 AM - 10 PM', location: '2nd Floor' },
    gym: { hours: '24 hours', location: '3rd Floor' },
  },
  hoursOfOperation: { weekday: '9 AM - 9 PM', weekend: '10 AM - 6 PM' },
  specialInstructions: ['Always mention the weekly wine tasting.'],
  welcomeMessage: 'Welcome to the Grand Hotel! How may I assist you?',
});

agent.run();

ReceptionistAgent

A front-desk agent that greets callers, collects their name and reason for calling, and transfers them to the appropriate department. Optionally supports visitor check-in as a TS-specific enhancement.

Source: src/prefabs/ReceptionistAgent.ts

Receptionist Configuration

The constructor accepts a ReceptionistConfig object:

Property	Type	Required	Default	Description
`name`	`string`	No	`"receptionist"`	Agent display name.
`route`	`string`	No	`"/receptionist"`	HTTP route for this agent.
`departments`	`ReceptionistDepartment[]`	Yes	--	Departments the agent can transfer callers to.
`greeting`	`string`	No	`"Thank you for calling. How can I help you today?"`	Initial greeting message.
`voice`	`string`	No	`"rime.spore"`	Voice identifier passed to `addLanguage`.
`companyName`	`string`	No	--	Optional company name appended to the greeting and used as a speech hint.
`checkInEnabled`	`boolean`	No	`false`	Whether the TS-specific `check_in_visitor` tool is registered.
`onVisitorCheckIn`	`(visitor: Record<string, string>) => void \| Promise<void>`	No	--	Callback fired when a visitor checks in.
`agentOptions`	`Partial<AgentOptions>`	No	--	Additional AgentBase options.

Each ReceptionistDepartment has the following shape:

Property	Type	Required	Description
`name`	`string`	Yes	Department identifier (used as enum value in `transfer_call`).
`description`	`string`	Yes	Description of the department (shown to the AI).
`number`	`string`	Yes	Phone number (or SIP address) to dial on transfer.

Receptionist Tools

`collect_caller_info`

Records the caller's name and reason for calling in global_data.caller_info via set_global_data.

Parameters:

Name	Type	Required	Description
`name`	`string`	Yes	The caller's name.
`reason`	`string`	Yes	The reason for the call.

Returns: An acknowledgement referencing the caller by name.

`transfer_call`

Transfers the caller to the selected department.

Parameters:

Name	Type	Required	Description
`department`	`string` (enum over department names)	Yes	The department to transfer to.

Behavior:

Uses post_process=true so the AI speaks the response before executing the transfer.
Uses FunctionResult.connect(number, final=true) to make the transfer permanent.
Reads global_data.caller_info.name (populated by collect_caller_info) to personalize the hand-off message.
Returns an error if the department name is unknown.

`check_in_visitor` (TS enhancement)

Only registered when checkInEnabled is true. Records a visitor in the per-call session and fires the onVisitorCheckIn callback.

Parameters:

Name	Type	Required	Description
`visitor_name`	`string`	Yes	Full name of the visitor.
`purpose`	`string`	Yes	Purpose of the visit.
`visiting`	`string`	Yes	Name of the person or department the visitor is here to see.

Receptionist Example

import { ReceptionistAgent } from '@signalwire/sdk';

const agent = new ReceptionistAgent({
  companyName: 'Acme Corporation',
  greeting: 'Thank you for calling Acme Corporation. How can I help you today?',
  voice: 'rime.spore',
  checkInEnabled: true,
  departments: [
    {
      name: 'engineering',
      description: 'Software development and technical teams',
      number: '+15551001001',
    },
    {
      name: 'hr',
      description: 'Employee services, benefits, and recruiting',
      number: '+15551001002',
    },
    {
      name: 'marketing',
      description: 'Brand, communications, and events',
      number: '+15551001003',
    },
    {
      name: 'executive',
      description: 'CEO and executive leadership',
      number: '+15551001004',
    },
  ],
  onVisitorCheckIn: async (visitor) => {
    console.log('Visitor checked in:', visitor);
    // Send Slack notification, update visitor log, print badge, etc.
  },
  agentOptions: {
    route: '/reception',
  },
});

agent.run();

Factory Functions

Each prefab provides a factory function that creates and returns a new instance. These are simple shorthands for new PrefabAgent(config):

Factory Function	Creates
`createInfoGathererAgent(config)`	`InfoGathererAgent`
`createSurveyAgent(config)`	`SurveyAgent`
`createFAQBotAgent(config)`	`FAQBotAgent`
`createConciergeAgent(config)`	`ConciergeAgent`
`createReceptionistAgent(config)`	`ReceptionistAgent`

Factory functions accept the same config type as their corresponding class constructors.

import { SurveyAgent } from '@signalwire/sdk';

const agent = new SurveyAgent({
  surveyName: 'Quick Survey',
  questions: [
    { id: 'q1', text: 'How was your experience?', type: 'rating' },
  ],
  onComplete: (responses, score) => {
    console.log('Done!', responses, score);
  },
});

agent.run();

The prefab classes (InfoGathererAgent, SurveyAgent, FAQBotAgent, ConciergeAgent, ReceptionistAgent) and their config types are exported from the main SDK entry point @signalwire/sdk. The create*Agent() factory helpers are defined alongside each prefab class in src/prefabs/; the class constructor shown above is the recommended public entry point.

Subclassing Prefabs

Prefab agents can be subclassed to customize their behavior. The two primary extension points are the static PROMPT_SECTIONS array and the defineTools() method.

Overriding `PROMPT_SECTIONS`

Each prefab declares a static override PROMPT_SECTIONS array that the AgentBase constructor merges into the prompt. You can override this in your subclass to change the AI's role and rules:

import { InfoGathererAgent } from '@signalwire/sdk';
import type { InfoGathererConfig } from '@signalwire/sdk';

class SpanishInfoGatherer extends InfoGathererAgent {
  static override PROMPT_SECTIONS = [
    {
      title: 'Role',
      body: 'You are a bilingual information-gathering assistant. You speak Spanish and English. Always respond in the same language the caller uses.',
    },
    {
      title: 'Rules',
      bullets: [
        'Ask one question at a time, in the order provided by start_questions.',
        'Use submit_answer to record each response and advance to the next question.',
        'If a question has confirm=true, insist on confirmation before submitting.',
        'If validation fails, politely ask the caller to try again in their language.',
      ],
    },
  ];

  constructor(config: InfoGathererConfig) {
    super(config);
  }
}

Overriding `defineTools()`

You can override defineTools() to add additional tools, replace existing tools, or extend the default tool set by calling super.defineTools() first:

import { FAQBotAgent, FunctionResult } from '@signalwire/sdk';
import type { FAQBotConfig } from '@signalwire/sdk';

class FAQBotWithFeedback extends FAQBotAgent {
  protected override defineTools(): void {
    // Register all the default tools first
    super.defineTools();

    // Add a custom feedback tool
    this.defineTool({
      name: 'submit_feedback',
      description: 'Allow the caller to submit feedback about the FAQ answers.',
      parameters: {
        type: 'object',
        properties: {
          rating: {
            type: 'string',
            description: 'How helpful was the answer: good, okay, or poor.',
          },
          comment: {
            type: 'string',
            description: 'Optional comment from the caller.',
          },
        },
        required: ['rating'],
      },
      handler: (args: Record<string, unknown>) => {
        const rating = args['rating'] as string;
        const comment = args['comment'] as string | undefined;
        console.log(`FAQ Feedback: ${rating}${comment ? ' - ' + comment : ''}`);
        return new FunctionResult('Thank you for your feedback!');
      },
    });
  }
}

Adding Dynamic Prompt Sections

Beyond the static PROMPT_SECTIONS, you can use this.promptAddSection() in the constructor body to add additional dynamic prompt content:

import { ConciergeAgent } from '@signalwire/sdk';
import type { ConciergeConfig } from '@signalwire/sdk';

class HolidayConcierge extends ConciergeAgent {
  constructor(config: ConciergeConfig) {
    super(config);

    // Add a holiday schedule section to the prompt
    this.promptAddSection('Holiday Schedule', {
      body: 'Note: The office is closed on December 25th and January 1st. All departments will reopen on January 2nd.',
    });
  }
}

Uh oh!

FilesExpand file tree

prefabs-guide.md

Latest commit

History

prefabs-guide.md

File metadata and controls

Prefab Agents Guide

Table of Contents

Overview

InfoGathererAgent

InfoGatherer Configuration

InfoGatherer Questions

InfoGatherer Tools

start_questions

submit_answer

Dynamic Mode

InfoGatherer Example

SurveyAgent

Survey Configuration

Survey Question Types

Survey Branching Logic

Survey Scoring

Survey Tools

validate_response

log_response

answer_question

get_current_question

get_survey_progress

Survey Example

FAQBotAgent

FAQBot Configuration

FAQBot Matching Engine

FAQBot Tools

search_faq

escalate

FAQBot Example

ConciergeAgent

Concierge Configuration

Concierge Tools

check_availability

get_directions

Concierge Example

ReceptionistAgent

Receptionist Configuration

Receptionist Tools

collect_caller_info

transfer_call

check_in_visitor (TS enhancement)

Receptionist Example

Factory Functions

Subclassing Prefabs

Overriding PROMPT_SECTIONS

Overriding defineTools()

Adding Dynamic Prompt Sections

`start_questions`

`submit_answer`

`validate_response`

`log_response`

`answer_question`

`get_current_question`

`get_survey_progress`

`search_faq`

`escalate`

`check_availability`

`get_directions`

`collect_caller_info`

`transfer_call`

`check_in_visitor` (TS enhancement)

Overriding `PROMPT_SECTIONS`

Overriding `defineTools()`