ElevenLabs Agents Platform

Overview

ElevenLabs Agents Platform is a comprehensive solution for building production-ready conversational AI voice agents. The platform coordinates four core components:

1. ASR (Automatic Speech Recognition) - Converts speech to text (32+ languages, sub-second latency)
2. LLM (Large Language Model) - Reasoning and response generation (GPT, Claude, Gemini, custom models)
3. TTS (Text-to-Speech) - Converts text to speech (5000+ voices, 31 languages, low latency)
4. Turn-Taking Model - Proprietary model that handles conversation timing and interruptions

🚨 Package Updates (November 2025)

ElevenLabs migrated to new scoped packages in August 2025:

DEPRECATED (Do not use):

• @11labs/react → DEPRECATED
• @11labs/client → DEPRECATED

Current packages:

npm install @elevenlabs/react@0.9.1        # React SDK
npm install @elevenlabs/client@0.9.1       # JavaScript SDK
npm install @elevenlabs/react-native@0.5.2 # React Native SDK
npm install @elevenlabs/elevenlabs-js@2.21.0 # Base SDK
npm install -g @elevenlabs/agents-cli@0.2.0  # CLI

If you have old packages installed, uninstall them first:

npm uninstall @11labs/react @11labs/client

When to Use This Skill

Use this skill when:

• Building voice-enabled customer support agents
• Creating interactive voice response (IVR) systems
• Developing conversational AI applications
• Integrating telephony (Twilio, SIP trunking)
• Implementing voice chat in web/mobile apps
• Configuring agents via CLI ("agents as code")
• Setting up RAG/knowledge bases for agents
• Integrating MCP (Model Context Protocol) servers
• Building HIPAA/GDPR-compliant voice systems
• Optimizing LLM costs with caching strategies

Platform Capabilities

Design & Configure:

• Multi-step workflows with visual builder
• System prompt engineering (6-component framework)
• 5000+ voices across 31 languages
• Pronunciation dictionaries (IPA/CMU formats)
• Speed control (0.7x-1.2x)
• RAG-powered knowledge bases
• Dynamic variables and personalization

Connect & Deploy:

• React SDK (@elevenlabs/react)
• JavaScript SDK (@elevenlabs/client)
• React Native SDK (@elevenlabs/react-native)
• Swift SDK (iOS/macOS)
• Embeddable widget
• Telephony integration (Twilio, SIP)
• Scribe (Real-Time Speech-to-Text) - Beta

Operate & Optimize:

• Automated testing (scenario, tool call, load)
• Conversation analysis and evaluation
• Analytics dashboard (resolution rates, sentiment, compliance)
• Privacy controls (GDPR, HIPAA, SOC 2)
• Cost optimization (LLM caching, model swapping, burst pricing)
• CLI for "agents as code" workflow

---

1. Quick Start (3 Integration Paths)

Path A: React SDK (Embedded Voice Chat)

For building voice chat interfaces in React applications.

Installation:

npm install @elevenlabs/react zod

Basic Example:

import { useConversation } from '@elevenlabs/react';
import { z } from 'zod';

export default function VoiceChat() {
  const { startConversation, stopConversation, status } = useConversation({
    // Public agent (no API key needed)
    agentId: 'your-agent-id',

    // OR private agent (requires API key)
    apiKey: process.env.NEXT_PUBLIC_ELEVENLABS_API_KEY,

    // OR signed URL (server-generated, most secure)
    signedUrl: '/api/elevenlabs/auth',

    // Client-side tools (browser functions)
    clientTools: {
      updateCart: {
        description: "Update the shopping cart",
        parameters: z.object({
          item: z.string(),
          quantity: z.number()
        }),
        handler: async ({ item, quantity }) => {
          console.log('Updating cart:', item, quantity);
          return { success: true };
        }
      }
    },

    // Event handlers
    onConnect: () => console.log('Connected'),
    onDisconnect: () => console.log('Disconnected'),
    onEvent: (event) => {
      switch (event.type) {
        case 'transcript':
          console.log('User said:', event.data.text);
          break;
        case 'agent_response':
          console.log('Agent replied:', event.data.text);
          break;
      }
    },

    // Regional compliance (GDPR, data residency)
    serverLocation: 'us' // 'us' | 'global' | 'eu-residency' | 'in-residency'
  });

  return (
    <div>
      <button onClick={startConversation}>Start Conversation</button>
      <button onClick={stopConversation}>Stop</button>
      <p>Status: {status}</p>
    </div>
  );
}

Path B: CLI ("Agents as Code")

For managing agents via code with version control and CI/CD.

Installation:

npm install -g @elevenlabs/agents-cli
# or
pnpm install -g @elevenlabs/agents-cli

Workflow:

# 1. Authenticate
elevenlabs auth login

# 2. Initialize project (creates agents.json, tools.json, tests.json)
elevenlabs agents init

# 3. Create agent from template
elevenlabs agents add "Support Agent" --template customer-service

# 4. Configure in agent_configs/support-agent.json

# 5. Push to platform
elevenlabs agents push --env dev

# 6. Test
elevenlabs agents test "Support Agent"

# 7. Deploy to production
elevenlabs agents push --env prod

Project Structure Created:

your_project/
├── agents.json              # Agent registry
├── tools.json               # Tool configurations
├── tests.json               # Test configurations
├── agent_configs/           # Individual agent files
├── tool_configs/            # Tool configuration files
└── test_configs/            # Test configuration files

Path C: API (Programmatic Agent Management)

For creating agents dynamically (multi-tenant, SaaS platforms).

Installation:

npm install elevenlabs

Example:

import { ElevenLabsClient } from 'elevenlabs';

const client = new ElevenLabsClient({
  apiKey: process.env.ELEVENLABS_API_KEY
});

// Create agent
const agent = await client.agents.create({
  name: 'Support Bot',
  conversation_config: {
    agent: {
      prompt: {
        prompt: "You are a helpful customer support agent.",
        llm: "gpt-4o",
        temperature: 0.7
      },
      first_message: "Hello! How can I help you today?",
      language: "en"
    },
    tts: {
      model_id: "eleven_turbo_v2_5",
      voice_id: "your-voice-id"
    }
  }
});

console.log('Agent created:', agent.agent_id);

---

2. Agent Configuration

System Prompt Architecture (6 Components)

ElevenLabs recommends structuring agent prompts using 6 components:

1. Personality

Define the agent's identity, role, and character traits.

Example:

You are Alex, a friendly and knowledgeable customer support specialist at TechCorp.
You have 5 years of experience helping customers solve technical issues.
You're patient, empathetic, and always maintain a positive attitude.

2. Environment

Describe the communication context (phone, web chat, video call).

Example:

You're speaking with customers over the phone. Communication is voice-only.
Customers may have background noise or poor connection quality.
Speak clearly and occasionally use thoughtful pauses for emphasis.

3. Tone

Specify formality, speech patterns, humor, and verbosity.

Example:

Tone: Professional yet warm. Use contractions ("I'm" instead of "I am") to sound natural.
Avoid jargon unless the customer uses it first. Keep responses concise (2-3 sentences max).
Use encouraging phrases like "I'll be happy to help with that" and "Let's get this sorted for you."

4. Goal

Define objectives and success criteria.

Example:

Primary Goal: Resolve customer technical issues on the first call.
Secondary Goals:
- Verify customer identity securely
- Document issue details accurately
- Offer proactive solutions
- End calls with confirmation that the issue is resolved

Success Criteria: Customer verbally confirms their issue is resolved.

5. Guardrails

Set boundaries, prohibited topics, and ethical constraints.

Example:

Guardrails:
- Never provide medical, legal, or financial advice
- Do not share confidential company information
- If asked about competitors, politely redirect to TechCorp's offerings
- Escalate to a human supervisor if customer becomes abusive
- Never make promises about refunds or credits without verification

6. Tools

Describe available external capabilities and when to use them.

Example:

Available Tools:
1. lookup_order(order_id) - Fetch order details from database. Use when customer mentions an order number.
2. transfer_to_supervisor() - Escalate to human agent. Use when issue requires manager approval.
3. send_password_reset(email) - Trigger password reset email. Use when customer can't access account.

Always explain to the customer what you're doing before calling a tool.

Complete Template:

{
  "agent": {
    "prompt": {
      "prompt": "Personality:\nYou are Alex, a friendly customer support specialist.\n\nEnvironment:\nYou're speaking with customers over the phone.\n\nTone:\nProfessional yet warm. Keep responses concise.\n\nGoal:\nResolve technical issues on the first call.\n\nGuardrails:\n- Never provide medical/legal/financial advice\n- Escalate abusive customers\n\nTools:\n- lookup_order(order_id) - Fetch order details\n- transfer_to_supervisor() - Escalate to human",
      "llm": "gpt-4o",
      "temperature": 0.7,
      "max_tokens": 500
    }
  }
}

Turn-Taking Modes

Controls when the agent interrupts or waits for the user to finish speaking.

3 Modes:

Mode	Behavior	Best For
Eager	Responds quickly, jumps in at earliest opportunity	Fast-paced support, quick orders
Normal	Balanced, waits for natural conversation breaks	General customer service (default)
Patient	Waits longer, allows detailed user responses	Information collection, therapy, tutoring

Configuration:

{
  "conversation_config": {
    "turn": {
      "mode": "patient" // "eager" | "normal" | "patient"
    }
  }
}

Use Cases:

• Eager: Fast food ordering, quick FAQs, urgent notifications
• Normal: General support, product inquiries, appointment booking
• Patient: Detailed form filling, emotional support, educational tutoring

Gotchas:

• Eager mode can feel interruptive to some users
• Patient mode may feel slow in fast-paced contexts
• Can be dynamically adjusted in workflows for context-aware behavior

Workflows (Visual Builder)

Create branching conversation flows with subagent nodes and conditional routing.

Node Types:

1. Subagent Nodes - Override base agent config (change prompt, voice, turn-taking)
2. Tool Nodes - Guarantee tool execution (unlike tools in subagents)

Configuration:

{
  "workflow": {
    "nodes": [
      {
        "id": "node_1",
        "type": "subagent",
        "config": {
          "system_prompt": "You are now a technical support specialist. Ask detailed diagnostic questions.",
          "turn_eagerness": "patient",
          "voice_id": "tech_support_voice_id"
        }
      },
      {
        "id": "node_2",
        "type": "tool",
        "tool_name": "transfer_to_human"
      }
    ],
    "edges": [
      {
        "from": "node_1",
        "to": "node_2",
        "condition": "user_requests_escalation"
      }
    ]
  }
}

Use Cases:

• Multi-department routing (sales → support → billing)
• Decision trees ("press 1 for sales, 2 for support")
• Role-playing scenarios (customer vs agent voices)
• Escalation paths (bot → human transfer)

Gotchas:

• Workflows add ~100-200ms latency per node transition
• Tool nodes guarantee execution (subagents may skip tools)
• Edges can create infinite loops if not tested properly

Dynamic Variables & Personalization

Inject runtime data into prompts, first messages, and tool parameters using {{var_name}} syntax.

System Variables (Auto-Available):

{{system__agent_id}}         // Current agent ID
{{system__conversation_id}}  // Conversation ID
{{system__caller_id}}        // Phone number (telephony only)
{{system__called_number}}    // Called number (telephony only)
{{system__call_duration_secs}} // Call duration
{{system__time_utc}}         // Current UTC time
{{system__call_sid}}         // Twilio call SID (Twilio only)

Custom Variables:

// Provide when starting conversation
const conversation = await client.conversations.create({
  agent_id: "agent_123",
  dynamic_variables: {
    user_name: "John",
    account_tier: "premium",
    order_id: "ORD-12345"
  }
});

Secret Variables (For API Keys):

{{secret__stripe_api_key}}
{{secret__database_password}}

Important: Secret variables only used in headers, never sent to LLM providers.

Usage in Prompts:

{
  "agent": {
    "prompt": {
      "prompt": "You are helping {{user_name}}, a {{account_tier}} customer."
    },
    "first_message": "Hello {{user_name}}! I see you're calling about order {{order_id}}."
  }
}

Gotcha: Missing variables cause "Missing required dynamic variables" error. Always provide all referenced variables when starting conversation.

Authentication Patterns

Option 1: Public Agents (No API Key)

const { startConversation } = useConversation({
  agentId: 'your-public-agent-id' // Anyone can use
});

Option 2: Private Agents with API Key

const { startConversation } = useConversation({
  agentId: 'your-private-agent-id',
  apiKey: process.env.NEXT_PUBLIC_ELEVENLABS_API_KEY
});

⚠️ Warning: Never expose API keys in client-side code. Use signed URLs instead.

Option 3: Signed URLs (Recommended for Production)

// Server-side (Next.js API route)
import { ElevenLabsClient } from 'elevenlabs';

export async function POST(req: Request) {
  const client = new ElevenLabsClient({
    apiKey: process.env.ELEVENLABS_API_KEY // Server-side only
  });

  const signedUrl = await client.convai.getSignedUrl({
    agent_id: 'your-agent-id'
  });

  return Response.json({ signedUrl });
}

// Client-side
const { startConversation } = useConversation({
  agentId: 'your-agent-id',
  signedUrl: await fetch('/api/elevenlabs/auth').then(r => r.json()).then(d => d.signedUrl)
});

---

3. Voice & Language Features

Multi-Voice Support

Dynamically switch between different voices during a single conversation.

Use Cases:

• Multi-character storytelling (different voice per character)
• Language tutoring (native speaker voices for each language)
• Role-playing scenarios (customer vs agent)
• Emotional agents (different voices for different moods)

Configuration:

{
  "agent": {
    "prompt": {
      "prompt": "When speaking as the customer, use voice_id 'customer_voice_abc123'. When speaking as the agent, use voice_id 'agent_voice_def456'."
    }
  }
}

Gotchas:

• Voice switching adds ~200ms latency per switch
• Requires careful prompt engineering to trigger switches correctly
• Not all voices work equally well for all characters

Pronunciation Dictionary

Customize how the agent pronounces specific words or phrases.

Supported Formats:

• IPA (International Phonetic Alphabet)
• CMU (Carnegie Mellon University Pronouncing Dictionary)
• Word Substitutions (replace words before TTS)

Configuration:

{
  "pronunciation_dictionary": [
    {
      "word": "ElevenLabs",
      "pronunciation": "ɪˈlɛvənlæbz",
      "format": "ipa"
    },
    {
      "word": "API",
      "pronunciation": "ey-pee-ay",
      "format": "cmu"
    },
    {
      "word": "AI",
      "substitution": "artificial intelligence"
    }
  ]
}

Use Cases:

• Brand names (e.g., "IKEA" → "ee-KAY-uh")
• Acronyms (e.g., "API" → "A-P-I" or "ay-pee-eye")
• Technical terms
• Character names in storytelling

Gotcha: Only Turbo v2/v2.5 models support phoneme-based pronunciation. Other models silently skip phoneme entries but still process word substitutions.

Speed Control

Adjust speaking speed dynamically (0.7x - 1.2x).

Configuration:

{
  "voice_settings": {
    "speed": 1.0 // 0.7 = slow, 1.0 = normal, 1.2 = fast
  }
}

Use Cases:

• Slow (0.7x-0.9x): Accessibility, children, non-native speakers
• Normal (1.0x): Default for most use cases
• Fast (1.1x-1.2x): Urgent notifications, power users

Best Practices:

• Use 0.9x-1.1x for natural-sounding adjustments
• Extreme values (below 0.7 or above 1.2) degrade quality
• Speed can be adjusted per agent, not per utterance

Voice Design

Create custom voices using ElevenLabs Voice Design tool.

Workflow:

1. Navigate to Voice Library → Create Voice
2. Use Voice Design (text-to-voice) or Voice Cloning (sample audio)
3. Test voice with sample text
4. Save voice to library
5. Use voice_id in agent configuration

Voice Cloning Best Practices:

• Use clean audio samples (no background noise, music, or pops)
• Maintain consistent microphone distance
• Avoid extreme volumes (whispering or shouting)
• 1-2 minutes of audio recommended

Gotcha: Using English-trained voices for non-English languages causes pronunciation issues. Always use language-matched voices.

Language Configuration

Support for 32+ languages with automatic detection and in-conversation switching.

Configuration:

{
  "agent": {
    "language": "en" // ISO 639-1 code
  }
}

Multi-Language Presets (Different Voice Per Language):

{
  "conversation_config": {
    "language_presets": [
      {
        "language": "en",
        "voice_id": "en_voice_id",
        "first_message": "Hello! How can I help you today?"
      },
      {
        "language": "es",
        "voice_id": "es_voice_id",
        "first_message": "¡Hola! ¿Cómo puedo ayudarte hoy?"
      },
      {
        "language": "fr",
        "voice_id": "fr_voice_id",
        "first_message": "Bonjour! Comment puis-je vous aider aujourd'hui?"
      }
    ]
  }
}

Automatic Language Detection: Agent detects user's language and switches automatically.

Supported Languages: English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Arabic, Chinese, Japanese, Korean, Hindi, and 18+ more.

---

4. Knowledge Base & RAG

RAG (Retrieval-Augmented Generation)

Enable agents to access large knowledge bases without loading entire documents into context.

How It Works:

1. Upload documents (PDF, TXT, DOCX) to knowledge base
2. ElevenLabs automatically computes vector embeddings
3. During conversation, relevant chunks retrieved based on semantic similarity
4. LLM uses retrieved context to generate responses

Configuration:

{
  "agent": {
    "prompt": {
      "knowledge_base": ["doc_id_1", "doc_id_2"]
    }
  }
}

Upload Documents via API:

import { ElevenLabsClient } from 'elevenlabs';

const client = new ElevenLabsClient({ apiKey: process.env.ELEVENLABS_API_KEY });

// Upload document
const doc = await client.knowledgeBase.upload({
  file: fs.createReadStream('support_docs.pdf'),
  name: 'Support Documentation'
});

// Compute RAG index
await client.knowledgeBase.computeRagIndex({
  document_id: doc.id,
  embedding_model: 'e5_mistral_7b' // or 'multilingual_e5_large'
});

Retrieval Configuration:

{
  "knowledge_base_config": {
    "max_chunks": 5,              // Number of chunks to retrieve
    "vector_distance_threshold": 0.8  // Similarity threshold
  }
}

Use Cases:

• Product documentation agents
• Customer support (FAQ, help center)
• Educational tutors (textbooks, lecture notes)
• Healthcare assistants (medical guidelines)

Gotchas:

• RAG adds ~500ms latency per query
• More chunks = higher cost but better context
• Higher vector distance = more context but potentially less relevant
• Documents must be indexed before use (can take minutes for large docs)

---

5. Tools (4 Types)

ElevenLabs supports 4 distinct tool types, each with different execution patterns.

A. Client Tools

Execute operations on the client side (browser or mobile app).

Use Cases:

• Update UI elements (shopping cart, notifications)
• Trigger navigation (redirect user to page)
• Access local storage
• Control media playback

React Example:

import { useConversation } from '@elevenlabs/react';
import { z } from 'zod';

const { startConversation } = useConversation({
  clientTools: {
    updateCart: {
      description: "Update the shopping cart with new items",
      parameters: z.object({
        item: z.string().describe("The item name"),
        quantity: z.number().describe("Quantity to add")
      }),
      handler: async ({ item, quantity }) => {
        // Client-side logic
        const cart = getCart();
        cart.add(item, quantity);
        updateUI(cart);
        return { success: true, total: cart.total };
      }
    },
    navigate: {
      description: "Navigate to a different page",
      parameters: z.object({
        url: z.string().describe("The URL to navigate to")
      }),
      handler: async ({ url }) => {
        window.location.href = url;
        return { success: true };
      }
    }
  }
});

Gotchas:

• Tool names are case-sensitive
• Must return a value (agent reads the return value)
• Handler can be async

B. Server Tools (Webhooks)

Make HTTP requests to external APIs from ElevenLabs servers.

Use Cases:

• Fetch real-time data (weather, stock prices)
• Update CRM systems (Salesforce, HubSpot)
• Process payments (Stripe, PayPal)
• Send emails/SMS (SendGrid, Twilio)

Configuration via CLI:

elevenlabs tools add-webhook "Get Weather" --config-path tool_configs/get-weather.json

tool_configs/get-weather.json:

{
  "name": "get_weather",
  "description": "Fetch current weather for a city",
  "url": "https://api.weather.com/v1/current",
  "method": "GET",
  "parameters": {
    "type": "object",
    "properties": {
      "city": {
        "type": "string",
        "description": "The city name (e.g., 'London', 'New York')"
      }
    },
    "required": ["city"]
  },
  "headers": {
    "Authorization": "Bearer {{secret__weather_api_key}}"
  }
}

Dynamic Variables in Tools:

{
  "url": "https://api.crm.com/customers/{{user_id}}",
  "headers": {
    "X-API-Key": "{{secret__crm_api_key}}"
  }
}

Gotchas:

• Secret variables only work in headers (not URL or body)
• Schema description guides LLM on when to use tool

C. MCP Tools (Model Context Protocol)

Connect to external MCP servers for standardized tool access.

Use Cases:

• Access databases (PostgreSQL, MongoDB)
• Query knowledge bases (Pinecone, Weaviate)
• Integrate with IDEs (VS Code, Cursor)
• Connect to data sources (Google Drive, Notion)

Configuration:

1. Navigate to MCP server integrations in dashboard
2. Click "Add Custom MCP Server"
3. Configure:
   • Name: Server identifier
   • Server URL: SSE or HTTP endpoint
   • Secret Token: Optional auth header
4. Test connectivity and discover tools
5. Add to agents (public or private)

Approval Modes:

• Always Ask: Maximum security, requires permission per tool call
• Fine-Grained: Per-tool approval settings
• No Approval: Auto-execute all tools

Gotchas:

• Only SSE and HTTP streamable transport supported
• MCP servers must be publicly accessible or behind auth
• Not available for Zero Retention Mode
• Not compatible with HIPAA compliance

Example: Using ElevenLabs MCP Server in Claude Desktop:

{
  "mcpServers": {
    "ElevenLabs": {
      "command": "uvx",
      "args": ["elevenlabs-mcp"],
      "env": {
        "ELEVENLABS_API_KEY": "<your-key>",
        "ELEVENLABS_MCP_OUTPUT_MODE": "files"
      }
    }
  }
}

D. System Tools

Modify the internal state of the conversation without external calls.