> ## 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. # OpenAI Models Integration > Configure GPT-4, GPT-3.5 Turbo, and other OpenAI models. Use OpenAI's powerful language models in your CallIntel agents. This guide covers setup, model selection, and best practices. ## Available Models ### GPT-4 (Recommended for Quality) **Model**: `gpt-4` or `gpt-4-turbo` ``` Capabilities: - Advanced reasoning - Complex problem-solving - Multi-step thinking - Best for critical tasks Cost: ~$0.03-0.06 per 1K tokens Speed: 1-3 seconds response time Best For: Customer service, technical support, complex queries ``` **When to Use GPT-4:** * Legal or compliance-critical interactions * Technical troubleshooting * Complex negotiations * High-value customer interactions ### GPT-3.5 Turbo (Recommended for Balance) **Model**: `gpt-3.5-turbo` ``` Capabilities: - Fast responses (< 1 second) - Good reasoning - Cost-effective - Excellent for most use cases Cost: ~$0.0005-0.002 per 1K tokens Speed: < 1 second response time Best For: General customer service, sales, lead generation ``` **When to Use GPT-3.5 Turbo:** * High-volume calling * Time-sensitive responses * Cost-conscious operations * Standard customer interactions ## Setup ### Step 1: Get Your API Key 1. Go to [platform.openai.com](https://platform.openai.com) 2. Sign in to your OpenAI account 3. Click "API keys" in the left menu 4. Click "Create new secret key" 5. Copy the key (you won't see it again) ``` Example Key Format: sk-proj-XXXXXxxxxxxxxxxxxxxxxxxxx ``` ### Step 2: Set up Billing OpenAI charges per API usage: 1. Go to "Billing" in OpenAI dashboard 2. Set usage limits to prevent surprises 3. Add payment method 4. Review pricing page for current rates ### Step 3: Add to CallIntel **For Super Admins:** 1. Go to Settings → Developer Settings 2. Click "API Keys" 3. Select "OpenAI" from provider list 4. Paste API key 5. Click "Test Connection" 6. Save **For Organization Admins:** 1. Go to Settings → AI Models 2. Click "Add OpenAI Model" 3. Paste API key (or use super admin configured key) 4. Select which models to enable 5. Save ### Step 4: Configure Agent 1. Create or Edit an Agent 2. Under "Language Model", select: * `gpt-4-turbo` or * `gpt-3.5-turbo` 3. Configure advanced settings: * **Temperature**: 0.7 (default) * **Max Tokens**: 150 (for responses) * **Frequency Penalty**: 0 (no penalty) * **Presence Penalty**: 0 (no penalty) 4. Save agent ### Step 5: Test Make a test call to verify: 1. Use Web Call feature 2. Speak to agent 3. Verify proper responses 4. Check call logs ## Model Selection ### Model Comparison | Feature | GPT-4 | GPT-3.5 | | --------- | ----- | ------- | | Quality | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | | Speed | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | | Cost | ⭐⭐ | ⭐⭐⭐⭐⭐ | | Reasoning | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | | Languages | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | ### Cost Example (1000 calls/month) **GPT-4 Scenario:** ``` 1000 calls × 500 tokens average × $0.00003/token = $15/month Cost per call: $0.015 Monthly cost: $15 (for 1000 calls) ``` **GPT-3.5 Turbo Scenario:** ``` 1000 calls × 500 tokens average × $0.0000005/token = $0.25/month Cost per call: $0.00025 Monthly cost: $0.25 (for 1000 calls) ``` **60x cost difference for 1000 calls!** ## Configuration Options ### Temperature Controls response creativity (0-2): ``` 0.0 = Deterministic (same response every time) 0.7 = Balanced (default - good for most uses) 1.5 = Creative (varied responses) 2.0 = Very creative (unpredictable) ``` **Recommended Settings:** ``` Customer Service: 0.3-0.5 (consistent) Sales: 0.7-0.9 (engaging but helpful) Creative Tasks: 1.2-1.5 (varied responses) ``` ### Max Tokens Maximum response length (1-4096): ``` Short Responses: 50-100 tokens Medium: 150-200 tokens Long Responses: 300-500 tokens ``` **Tip**: Lower max tokens saves costs! ``` Max Tokens Example: 50 tokens: $0.0015 per call 150 tokens: $0.0045 per call 300 tokens: $0.009 per call ``` ### System Prompt Define agent behavior (see [Agent Setup Guide](../core-concepts/agents)): ``` Example: "You are a friendly customer service representative. Keep responses under 100 words. Be helpful and professional." ``` ## Advanced Features ### Function Calling Enable agents to call external functions: ```json theme={null} { "name": "get_order_status", "description": "Get status of customer order", "parameters": { "type": "object", "properties": { "order_id": { "type": "string", "description": "The order ID" } } } } ``` Agents can: * Look up information * Process transactions * Update systems * Trigger actions ### Vision Capabilities GPT-4V can analyze images: ``` Supported: - Receipt analysis - Document scanning - Image description - Quality inspection ``` **Note**: Requires image input in calls (enterprise feature). ## Cost Optimization ### 1. Use GPT-3.5 Turbo for Most Tasks ``` GPT-4: Use sparingly for complex queries GPT-3.5: Default for all other interactions Savings: 10-20x cost reduction ``` ### 2. Optimize Prompts **Bad (expensive):** ``` "You are a helpful AI assistant in a contact center. You should be friendly, professional, and knowledgeable about our products. When customers call, listen carefully to their questions and provide helpful, accurate responses..." ``` **Good (cheaper):** ``` "You are a helpful customer service representative. Be friendly and professional." ``` **Savings**: 60% reduction in tokens! ### 3. Reduce Token Usage ``` Technique | Savings -----------------|------- Shorter KB | 30-50% Concise prompts | 20-40% Lower max_tokens | 10-30% No history | 10-20% Combined: 50-70% possible ``` ### 4. Batch Processing For scheduled calls, use batch API: ``` Standard: $0.002 per 1K tokens Batch: $0.0005 per 1K tokens Savings: 75% cheaper ``` **How to Use:** 1. Schedule calls for off-peak hours 2. Use batch endpoint (CallIntel handles this) 3. 24-hour processing window 4. Save significantly on costs ## Monitoring & Limits ### Token Monitoring Check usage in OpenAI dashboard: 1. Go to [Usage Dashboard](https://platform.openai.com/account/usage/overview) 2. View current usage 3. Check spending 4. View by model breakdown ### Rate Limits Default limits for standard accounts: ``` GPT-4: 200 requests/minute GPT-3.5: 3,500 requests/minute Increase limits in account settings Contact OpenAI for higher limits ``` ### Budget Controls Set spending limits to prevent surprises: 1. Go to Billing → Usage Limits 2. Set hard limit (e.g., \$100/month) 3. Optional email alert at 50% 4. API requests blocked when limit reached ## Best Practices ### 1. Start with GPT-3.5 Turbo ``` Phase 1: Use GPT-3.5 for all agents Phase 2: A/B test GPT-4 on specific agents Phase 3: Use GPT-4 only where needed Result: Optimized cost and quality balance ``` ### 2. Monitor Performance Track key metrics: ``` - Average response time - User satisfaction - Cost per call - Error rate - Token usage ``` Review monthly and optimize. ### 3. Test Before Production ``` 1. Create test agents with both models 2. Make identical calls 3. Compare responses 4. Evaluate cost/quality tradeoff 5. Choose best option for each use case ``` ### 4. Use Lower Temperature for Consistency ``` Customer Service: 0.3-0.5 - Consistent responses - Predictable behavior - Better for compliance-critical tasks ``` ## Troubleshooting Verify key starts with `sk-` and is not truncated. Copy from OpenAI dashboard again and test. Switch to GPT-3.5 Turbo for faster responses. Also reduce max\_tokens setting. Check token usage in OpenAI dashboard. Reduce max\_tokens, shorten KB, and simplify prompts. Lower temperature setting to 0.3-0.5 for more deterministic responses. Consider upgrading to GPT-4 for better reasoning, or improve system prompt and knowledge base. *** ## See Also Compare all AI providers Create and configure agents Build effective knowledge bases *** ## Support [View OpenAI Documentation](https://platform.openai.com/docs) Email: [callintel01@gmail.com](mailto:callintel01@gmail.com)