Creating an Agent

voice_list = {
  "gHu9GtaHOXcSqFTK06ux": {
    "language": "Hindi",
    "gender": "Female",
    "accent": "Standard",
    "name": "Anjali - Soothing Hindi Voice"
  },
  "m5qndnI7u4OAdXhH0Mr5": {
    "language": "Hindi",
    "gender": "Male",
    "accent": "Standard",
    "name": "Krishna - Energetic Hindi Voice"
  },
  "90ipbRoKi4CpHXvKVtl0": {
    "language": "English",
    "gender": "Female",
    "accent": "Indian",
    "name": "Anika - Customer Care Agent"
  },
  "xnx6sPTtvU635ocDt2j7": {
    "language": "English",
    "gender": "Male",
    "accent": "Indian",
    "name": "Chinmay - Calm, Energetic & Relatable"
  },
  "kdmDKE6EkgrWrrykO9Qt": {
    "language": "English",
    "gender": "Female",
    "accent": "American",
    "name": "Alexandra"
  },
  "XA2bIQ92TabjGbpO2xRr": {
    "language": "English",
    "gender": "Male",
    "accent": "American",
    "name": "Jerry"
  },
  "MzqUf1HbJ8UmQ0wUsx2p": {
    "language": "English",
    "gender": "Female",
    "accent": "British",
    "name": "Katie X"
  },
  "qxjGnozOAtD4eqNuXms4": {
    "language": "English",
    "gender": "Male",
    "accent": "British",
    "name": "John Shaw - Polite Customer Care Voice"
  },
  "6qpxBH5KUSDb40bij36w": {
    "language": "English",
    "gender": "Female",
    "accent": "Singaporean",
    "name": "Lilian"
  },
  "aCChyB4P5WEomwRsOKRh": {
    "language": "English",
    "gender": "Female",
    "accent": "Middle east",
    "name": "Salma - Conversational Expressive Voice"
  }
}

{
  "provider": "assembly-ai", // Required: transcription provider

  "confidenceThreshold": 0.4, // Optional: transcripts below this confidence will be discarded (>=0, <=1), default is 0.4

  "disablePartialTranscripts": false, // Optional: set to true to disable receiving partial transcripts, default is false

  "endUtteranceSilenceThreshold": 1500, // Optional: silence duration (in ms) that marks end of an utterance

  "fallbackPlan": {
    "provider": "backup-provider",
    "strategy": "retry" // Optional: example plan in case the main provider fails
  },

  "language": "en", // Optional: language for transcription, defaults to "en"

  "realtimeUrl": "wss://transcriber.example.com/socket", // Optional: WebSocket URL the transcriber connects to

  "wordBoost": ["blockchain", "quantum computing", "neural networks"] // Optional: custom vocabulary, up to 2500 characters
}

{
  "provider": "azure", // Required: transcription provider
  "language": "en-US" // Optional: transcription language (must be from Azure's supported list)
  // list of azure supported languages
  // https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=stt
}

{
  "provider": "azure", // Required: transcription provider
  "language": "te-IN" // Optional: transcription language (must be from Azure's supported list)
  // list of azure supported languages
  // https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=stt
}

{
  "provider": "azure", // Required: transcription provider
  "language": "ta-IN" // Optional: transcription language (must be from Azure's supported list)
  // list of azure supported languages
  // https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=stt
}

{
  // Required: deepgram transcriber config
  "provider": "deepgram",

  "codeSwitchingEnabled": false,
  // Optional: Enables automatic switching between supported languages mid-call.
  // Default is false.
  // Supported languages: https://developers.deepgram.com/docs/supported-languages

  "confidenceThreshold": 0.4,
  // Optional: Transcripts below this confidence will be discarded.
  // Must be between 0 and 1. Default is 0.4.

  "endpointing": 10,
  // Optional: Timeout (in ms) after user silence to send transcript.
  // Recommended values: 10 (faster, less accurate for short phrases), or 300 (slightly slower, more accurate).
  // Docs: https://developers.deepgram.com/docs/endpointing
  // Range: 10–500, default is 10.

  "fallbackPlan": {
    "provider": "backup-deepgram",
    "strategy": "failover"
  },
  // Optional: Fallback voice provider if Deepgram fails

  "keyterm": ["priority support", "platinum subscription"],
  // Optional: Keyterm Prompting improves Keyword Recall Rate (KRR)

  "keywords": ["VapiAI", "DeepSync", "EdgeCompute"],
  // Optional: Helps Deepgram model recognize uncommon or domain-specific terms

  "language": "en-US",
  // Optional: Set the transcription language
  // Supported languages: https://developers.deepgram.com/docs/models-languages-overview

  "mipOptOut": false,
  // Optional: Opt out of Deepgram's Model Improvement Program
  // Docs: https://developers.deepgram.com/docs/the-deepgram-model-improvement-partnership-program#want-to-opt-out

  "model": "nova",
  // Optional: Specifies the Deepgram model to use
  // Available models: https://developers.deepgram.com/docs/models-languages-overview

  "numerals": true,
  // Optional: Converts spoken numbers into numeric form
  // e.g., "nine-seven-two" becomes "972"

  "smartFormat": true
  // Optional: Applies smart formatting to transcription output
  // May format numbers as times; use with caution
}

{
  // Required: 11labs transcriber config
  "provider": "11labs",

  "fallbackPlan": {
    "provider": "backup-11labs",
    "strategy": "failover"
  },
  // Optional: Fallback voice provider if 11labs fails

  "language": "en",
  // Optional: Language for transcription
  // Must be one of the ISO 639-1 or ISO 639-2 language codes (e.g., "en", "fr", "es", "hi", "zh", "yue")
  // Full list of supported codes is extensive, including:
  // ar (Arabic), bn (Bengali), zh (Chinese), en (English), fr (French), de (German), hi (Hindi), ja (Japanese), ru (Russian), es (Spanish), etc.

  "model": "scribe_v1"
  // Optional: Transcription model to use
  // Currently defaults to "scribe_v1"
}

{
  // Required: OpenAI llm
  "provider": "openai",

  "model": "gpt-4o",
  // Required: The OpenAI model to use for transcription and conversation
  // Options include: "gpt-4o", "gpt-4", "gpt-4-turbo", "gpt-3.5-turbo", etc.
  // Full model list provided by Vapi

  "emotionRecognitionEnabled": false,
  // Optional: Whether to detect user emotion from speech
  // Default is false, as OpenAI models infer emotion well from text alone

  "fallbackModels": ["gpt-3.5-turbo", "gpt-4-0613"],
  // Optional: Models to fall back to if primary model fails
  // Vapi usually auto-selects optimal fallbacks, only override if necessary

  "knowledgeBase": {
    // Optional: Custom knowledge base support
    "server": {
      "url": "https://your-knowledge-api.com/query",
      // Required: API endpoint for your knowledge base

      "timeoutSeconds": 20,
      // Optional: Timeout for requests in seconds (1–120)
      // Default is 20

      "secret": "my-secret-key",
      // Optional: Secret sent as `x-vapi-secret` header

      "headers": {
        "Authorization": "Bearer custom-token",
        "Custom-Header": "value"
      },
      // Optional: Additional headers to send with the request

      "backoffPlan": {
        "initialDelay": 1000,
        "maxDelay": 10000,
        "multiplier": 2
      }
      // Optional: Retry strategy if requests to your knowledge base fail
    },

    "knowledgeBaseId": "kb-ocean-facts"
    // Optional: ID of the knowledge base to be used
  },

  "maxTokens": 500,
  // Optional: Max tokens the assistant can generate per turn (50–10000)
  // Default is 250

  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user",
      "content": "How does transcription work?"
    }
  ],
  // Optional: Initial conversation state provided to the model

  "numFastTurns": 2,
  // Optional: Use faster, smaller model (e.g., gpt-3.5) for first N turns
  // Default is 0

  "temperature": 0.7,
  // Optional: Controls randomness/creativity of the output (0–2)
  // Default is 0 for fast caching and low latency

  "toolIds": [
    "7709cdd9-4fff-4c6d-8d34-dc2e4988550a",
    "1ecd7247-a3c2-423f-8c5f-08b379c7b055"
  ]
  // Optional: List of tools the assistant can use during conversation
  // Can be used alongside `tools` for transient tool support
}

{
  // Required: The provider ID for this configuration
  "provider": "11labs",

  "voiceId": "FFmp1h1BMl0iVHA0JxrI",
  // Checkout Voices :  https://elevenlabs.io/app/voice-library
  // Required: Voice ID from your 11Labs Voice Library
  // Can be a string or enum — ensure it exists in your library

  "autoMode": false,
  // Optional: Enables or disables automatic voice settings mode
  // Default is false

  "cachingEnabled": true,
  // Optional: Enables voice caching for performance optimization
  // Default is true

  "chunkPlan": {
    "chunkSizeMs": 400,
    // Optional: Duration (in milliseconds) of each audio chunk

    "delayBeforeFirstChunkMs": 100,
    // Optional: Delay (in ms) before the first audio chunk is sent

    "maxChunkDelayMs": 300,
    // Optional: Max delay between chunks (in ms)

    "maxChunkSizeMs": 1000
    // Optional: Max allowed size of a chunk (in ms)
  },
  // Optional: Controls how text-to-speech output is broken into chunks

  "enableSsmlParsing": false,
  // Optional: Enables [SSML parsing](https://elevenlabs.io/docs/speech-synthesis/prompting#pronunciation)
  // Default is false to reduce latency

  "language": "en",
  // Optional: Enforced language (ISO 639-1 format)
  // Only supported in Turbo v2.5 models — others will return an error if set

  "model": "eleven_turbo_v2_5",
  // Optional: Voice model to be used
  // Available options:
  // "eleven_multilingual_v2", "eleven_turbo_v2", "eleven_turbo_v2_5",
  // "eleven_flash_v2", "eleven_flash_v2_5", "eleven_monolingual_v1"

  "optimizeStreamingLatency": 3,
  // Optional: Controls latency optimization (range: 0–4)
  // Default is 3

  "similarityBoost": 0.8,
  // Optional: Controls how closely the voice matches original training samples (0–1)

  "speed": 1.0,
  // Optional: Adjusts speech rate (range: 0.7–1.2)

  "stability": 0.6,
  // Optional: Controls variation vs consistency in voice generation (0–1)

  "style": 0.5,
  // Optional: Adds expressiveness/styling to the voice (0–1)

  "useSpeakerBoost": true
  // Optional: Enables speaker boost for clarity and energy
}

        {
            // Required: The provider ID for this configuration
            "provider": "11labs",

            "voiceId": "gCr8TeSJgJaeaIoV4RWH",

            # Remainging Other Values as mentioned above
        }

{
    // Required: The provider ID for this configuration
    "provider": "azure",

    "voiceId": "te-IN-MohanNeural",
    # Male Telugu Voice : te-IN-MohanNeural
    # Female Telugu Voice : te-IN-ShrutiNeural
}

{
    // Required: The provider ID for this configuration
    "provider": "azure",

    "voiceId": "ta-IN-PallaviNeural",
    # Female Tamil Voice : ta-IN-PallaviNeural
}