> ## Documentation Index
> Fetch the complete documentation index at: https://docs.callintel.io/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Models & Integrations

> Configure language models, voice synthesis, and speech recognition. CallIntel supports integration with leading AI platforms for language understanding, voice synthesis, and speech recognition. Choose the models and providers that best fit your needs.

## Supported Providers

### Language Models (LLMs)

CallIntel integrates with multiple LLM providers:

<CardGroup cols={2}>
  <Card title="OpenAI" icon="OpenAI" href="/guides/integrations/ai-models/openai">
    GPT-4, GPT-3.5 Turbo, and other models
  </Card>

  <Card title="Google Gemini" icon="Google" href="/guides/integrations/ai-models/gemini">
    Gemini Pro and vision-enabled models
  </Card>

  <Card title="Qwen (Alibaba)" icon="code" href="/guides/integrations/ai-models/qwen">
    Qwen models for cost-effective inference
  </Card>

  <Card title="Claude" icon="code" href="/guides/integrations/ai-models/claude">
    Anthropic's Claude models (coming soon)
  </Card>
</CardGroup>

### Voice Input (Speech-to-Text)

<CardGroup cols={3}>
  <Card title="AssemblyAI" icon="microphone">
    Real-time speech recognition
  </Card>

  <Card title="Deepgram" icon="microphone">
    Low-latency speech-to-text
  </Card>

  <Card title="Cartesia" icon="microphone">
    Enterprise speech recognition
  </Card>
</CardGroup>

### Voice Output (Text-to-Speech)

<CardGroup cols={3}>
  <Card title="OpenAI" icon="microphone">
    Natural voice synthesis
  </Card>

  <Card title="Cartesia" icon="microphone">
    High-quality voice output
  </Card>

  <Card title="ElevenLabs" icon="microphone">
    Realistic voice cloning
  </Card>

  <Card title="Rime AI" icon="microphone">
    Fast, natural synthesis
  </Card>

  <Card title="Inworld AI" icon="microphone">
    Interactive character voices
  </Card>
</CardGroup>

### Speaker Identification (Diarization)

<CardGroup cols={1}>
  <Card title="Pyannote" icon="users">
    Identify and distinguish speakers
  </Card>
</CardGroup>

***

## Model Selection Guide

### Choosing a Language Model

**For General Purpose:**

* **OpenAI GPT-4** - Best quality, highest cost
* **OpenAI GPT-3.5 Turbo** - Great quality, more affordable
* **Gemini Pro** - Good quality, competitive pricing

**For Cost-Effective:**

* **Qwen Models** - Lower cost, reasonable quality
* **GPT-3.5 Turbo** - Balance of quality and cost

**For Specific Tasks:**

* **Customer Service** - GPT-3.5 Turbo recommended
* **Technical Support** - GPT-4 recommended
* **Sales Calls** - Gemini Pro recommended
* **High Volume** - Qwen recommended

### Choosing a Speech-to-Text Provider

**For Real-Time Performance:**

* **Deepgram** - Lowest latency (200ms)
* **Cartesia** - Enterprise-grade performance
* **AssemblyAI** - Reliable and scalable

**For Accuracy:**

* **Deepgram Nova 2** - Best accuracy
* **Cartesia** - Domain-specific models
* **AssemblyAI** - Excellent general accuracy

**For Cost:**

* **AssemblyAI** - Competitive pricing
* **Deepgram** - Flexible pricing tiers
* **Cartesia** - Enterprise custom pricing

### Choosing a Text-to-Speech Provider

**For Natural Sound:**

* **OpenAI** - Most natural, good variety
* **ElevenLabs** - Premium quality voices
* **Cartesia** - Studio-quality audio

**For Speed:**

* **Rime AI** - Fastest synthesis
* **Cartesia** - Enterprise performance
* **OpenAI** - Good balance

**For Voice Variety:**

* **ElevenLabs** - Extensive voice library
* **OpenAI** - Multiple voices, multiple languages
* **Inworld AI** - Character-specific voices

***

## Configuration

### Setting Up Your API Keys

Each provider requires API credentials:

**Step 1: Obtain API Key**

* Visit provider's dashboard
* Create API key in settings
* Copy key securely

**Step 2: Add to CallIntel**

1. Go to Developer Settings
2. Select "API Keys" section
3. Paste provider API key
4. Save configuration

**Step 3: Test Connection**

* Click "Test" button
* Verify successful connection
* Save settings

### Organization-Level Setup

Super Admins configure providers globally:

```
1. Admin Dashboard → Settings
2. Select provider (OpenAI, Gemini, etc.)
3. Enter API key
4. Configure rate limits
5. Enable/disable provider
6. Save
```

### Agent-Level Configuration

Organizations select providers per agent:

```
1. Create/Edit Agent
2. Select Language Model (LLM)
3. Select Speech-to-Text (STT)
4. Select Text-to-Speech (TTS)
5. Save agent
```

***

## Quick Start

### Basic Setup (5 minutes)

<Steps>
  <Step title="Choose OpenAI">
    Start with OpenAI GPT-3.5 for best results
  </Step>

  <Step title="Get API Key">
    Visit [https://platform.openai.com/api-keys](https://platform.openai.com/api-keys)
  </Step>

  <Step title="Add to CallIntel">
    Settings → API Keys → Add OpenAI
  </Step>

  <Step title="Create Agent">
    Use OpenAI model in new agent
  </Step>

  <Step title="Make Test Call">
    Test agent with web call
  </Step>
</Steps>

***

## Cost Optimization

### Token Usage Tips

**Reduce Token Consumption:**

* Shorter knowledge bases (only essential info)
* Concise system prompts
* Lower max\_tokens setting
* Filter conversations to relevant context

**Example Cost Reduction:**

```
Before:
- Long KB: 5,000 tokens per call
- Verbose prompt: 2,000 tokens
- Total: 7,000 tokens × $0.002 = $0.014 per call

After:
- Optimized KB: 2,000 tokens
- Concise prompt: 500 tokens
- Total: 2,500 tokens × $0.002 = $0.005 per call

Savings: 65% reduction
```

### Batch Processing

Use batch endpoints for non-real-time calls:

* Batch pricing: 50% discount
* Perfect for scheduled campaigns
* Save costs on high-volume calls

### Model Selection Strategy

```
High Volume + Cost Sensitive:
→ Use Qwen models (lowest cost)

Quality Critical:
→ Use GPT-4 or Claude (best quality)

Balanced Approach:
→ Use GPT-3.5 Turbo (sweet spot)

Real-Time Performance:
→ Pair with Deepgram STT
```

***

## Detailed Provider Guides

<CardGroup cols={2}>
  <Card title="OpenAI Models" icon="openai" href="/guides/integrations/ai-models/openai">
    Complete guide to GPT-4, GPT-3.5, and more
  </Card>

  <Card title="Google Gemini" icon="google" href="/guides/integrations/ai-models/gemini">
    Integrate Google's Gemini models
  </Card>

  <Card title="Speech Recognition" icon="microphone" href="/guides/integrations/ai-models/speech-to-text">
    AssemblyAI, Deepgram, Cartesia setup
  </Card>

  <Card title="Voice Synthesis" icon="waveform" href="/guides/integrations/ai-models/text-to-speech">
    OpenAI, ElevenLabs, Cartesia voices
  </Card>

  <Card title="Speaker Diarization" icon="users" href="/guides/integrations/ai-models/speaker-identification">
    Identify and distinguish speakers
  </Card>

  <Card title="Overview" icon="star" href="/guides/integrations/ai-models/overview">
    Compare all AI providers
  </Card>
</CardGroup>

***

## Troubleshooting

<Accordion title="How do I switch between models?">
  Edit your agent configuration, select a different model from the dropdown, and save. Existing calls continue with old model; new calls use new model.
</Accordion>

<Accordion title="Why are my calls slow?">
  Check your STT provider latency, LLM response time, and TTS speed. Try switching to faster providers like Deepgram and Rime AI.
</Accordion>

<Accordion title="API key invalid error?">
  Verify API key is correct, not expired, and has proper permissions. Regenerate key if necessary.
</Accordion>

<Accordion title="How do I monitor token usage?">
  Check your provider's dashboard (OpenAI, Gemini, etc.) for real-time usage metrics and costs.
</Accordion>

***

## Support

<CardGroup cols={2}>
  <Card title="API Documentation" icon="book" href="api-reference/introduction">
    View complete provider documentation
  </Card>

  <Card title="Integration Help" icon="envelope">
    Email: [callintel01@gmail.com](mailto:callintel01@gmail.com)
  </Card>
</CardGroup>