autoch.at Documentation

Appearance

AI Configuration ​

Note: This documentation refers to the legacy AI Configuration system. The platform now uses the Assistants system for AI configuration.

Current Path: /assistants (AI Assistant → Assistants in navigation)
Legacy Path: /settings/ai (redirects to /assistants)

For the current AI configuration system, see:

Overview (Legacy System) ​

AI configuration determines:

  • How your AI assistant responds to customers
  • What information it collects
  • How it qualifies leads
  • The tone and personality of interactions

Settings apply across all channels (voice, SMS, chat, email) unless overridden by channel-specific configurations.

Migration Note: If you previously used the AI Configuration page, your settings have been migrated to the Assistants system. You can now create multiple assistants with different configurations for different scenarios.

Greeting Message ​

The first message your AI assistant says when starting a conversation. This sets the tone and helps customers understand what to expect.

Best Practices:

  • Keep it friendly and welcoming
  • Set clear expectations
  • Mention your company name
  • Be concise (1-2 sentences)

Example:

  • "Hi! Thanks for reaching out to [Company Name]. How can I help you today?"
  • "Welcome! I'm here to answer your questions about our services."

Note: This greeting is used across all channels unless overridden by channel-specific settings or messaging directives.

Voice Selection ​

Choose the voice used for phone calls and voice-enabled chat. Each voice has unique characteristics:

Available Voices:

  • Elliot (Male) - Professional, clear
  • Kylie (Female) - Friendly, warm
  • Rohan (Male) - Confident, authoritative
  • Lily (Female) - Energetic, upbeat
  • Savannah (Female) - Calm, reassuring
  • And many more...

Considerations:

  • Match voice to your brand personality
  • Consider your target audience
  • Test different voices to find the best fit
  • Voice can be previewed in voice settings

Default: Elliot

LLM Provider ​

Choose your AI model provider. Each provider offers different strengths:

OpenAI (GPT) ​

  • Best for: Speed and responsiveness
  • Strengths: Fast responses, good for high-volume
  • Models: GPT-4o, GPT-4o Mini, GPT-4 Turbo, GPT-4, GPT-3.5 Turbo
  • Cost: Moderate per token

Google (Gemini) ​

  • Best for: Cost-effectiveness
  • Strengths: Good balance of quality and cost
  • Models: Gemini Pro, Gemini Pro Vision
  • Cost: Lower per token

Recommendation: Start with OpenAI for fast, high-quality responses, or Google if cost is a primary concern.

Model Selection ​

Choose the specific AI model version. Higher-end models offer better quality but cost more.

OpenAI Models ​

GPT-4o Mini (Recommended)

  • Best balance of quality, speed, and cost
  • Fast and capable
  • Recommended for most use cases
  • Optimized for efficiency

GPT-4o

  • High quality with improved speed
  • Better performance than GPT-4
  • Good for complex tasks requiring accuracy

GPT-4 Turbo

  • Fast and capable
  • Good for high-volume scenarios
  • Balanced performance

GPT-4

  • High quality
  • Slower than newer models
  • Best for complex tasks requiring maximum accuracy

GPT-3.5 Turbo

  • Fast and cost-effective
  • Good for simple interactions
  • Lower quality than GPT-4 models

Google Models ​

Gemini Pro

  • Good quality
  • Cost-effective
  • Suitable for most use cases

Gemini Pro Vision

  • Includes image understanding
  • Use when processing visual content
  • Higher cost than Pro

System Prompt ​

Core instructions that define your AI's behavior, personality, and goals. This is the most important configuration setting.

Best Practices ​

  1. Be Specific:

    • Include your company name and industry
    • Describe your products/services
    • Define qualification criteria clearly

  2. Set Clear Goals:

    • What should the AI accomplish?
    • What information should it collect?
    • How should it qualify leads?

  3. Define Tone:

    • Professional vs. casual
    • Formal vs. friendly
    • Technical vs. simple

  4. Include Context:

    • Company background
    • Target customer profile
    • Common questions and answers

  5. Set Boundaries:

    • What the AI should NOT do
    • Topics to avoid
    • When to escalate to human

Example System Prompt ​

You are a helpful AI assistant for [Company Name], a [industry] company.

Your goals:
1. Answer questions about our [products/services]
2. Qualify leads by collecting: budget, timeline, project scope
3. Schedule appointments when requested
4. Be friendly, professional, and concise

Qualification criteria:
- Budget: Minimum $X
- Timeline: Within X months
- Project scope: [specific requirements]

Always:
- Use the customer's name when known
- Confirm important details
- Offer to schedule a call if appropriate

Never:
- Make promises we can't keep
- Discuss pricing without qualification
- Share confidential information

Using Templates ​

The system includes pre-built prompt templates organized by category:

  • Sales & Lead Qualification: Templates for qualifying leads and sales conversations
  • Customer Support: Templates for customer service and support interactions
  • Multi-Channel Messaging: Templates optimized for specific channels (Chat, Email, SMS)
  • Industry-Specific: Templates tailored to different industries and use cases
  • General Purpose: Versatile templates for common business scenarios

Template Categories:

  • Lead Qualification (SaaS, E-commerce, Services)
  • Customer Support (General, Technical, Billing)
  • Sales (Product Sales, Service Sales, Upselling)
  • Industry-Specific (Healthcare, Legal, Real Estate, etc.)

Using Templates:

  1. Navigate to AI Assistant → Assistants (or use the legacy path /settings/ai which redirects)
  2. Create or edit an assistant
  3. Scroll to the Prompt Templates section
  4. Browse templates by category
  5. Click Apply Template to use a template
  6. Customize the template for your specific needs

Note: All templates use {company_name} placeholder which is automatically replaced with your business name. Templates have been cleaned up and updated to remove hardcoded business names and duplicates.

Temperature ​

Controls response creativity and randomness. Lower values produce more consistent, factual responses. Higher values produce more creative, varied responses.

Range: 0.0 to 1.0

Recommended Values:

  • 0.3-0.5: More consistent, factual (best for qualification)
  • 0.7: Balanced (recommended for most use cases)
  • 0.8-0.9: More creative, varied (use sparingly)

When to Adjust:

  • Lower temperature: When you need consistent, accurate responses
  • Higher temperature: When you want more natural, varied conversations

Default: 0.7

Channel-Specific Overrides ​

While these settings apply globally, you can override them per channel using:

  • Messaging Directives: Channel-specific instructions
  • Assistants: Create channel-specific assistants
  • Routing Rules: Route to different assistants based on conditions

Migration to Messaging Directives ​

The platform is moving to a centralized messaging directive system. System prompts configured here will be migrated to the new system. For new configurations, consider using assistants and routing rules for channel-specific behavior.

AI Autonomy Controls ​

Hard Stop Conditions ​

Set up which intents will always block AI from sending automatically - your safety net:

Default Hard Stop Intents:

  • Billing Promise: Any promise or commitment related to billing
  • Legal Issue: Legal matters or concerns
  • Scope Change: Changes to project scope or requirements
  • Contract Terms: Discussion of contract terms or agreements

How It Works:

  • When AI detects a hard stop intent, auto-send is automatically blocked
  • Everything needs human review before sending
  • This applies no matter what your autonomy mode is set to
  • You can see all hard stop triggers in your audit logs

Configuring Hard Stops:

  1. Go to Settings → AI Autonomy (admin-only)
  2. See your current hard stop intents
  3. Add or remove intent types as needed
  4. Save your changes

Tips:

  • Keep the default hard stops enabled - they're there for your safety
  • Add more intents based on what your business needs
  • Review hard stop triggers in audit logs to see what's being caught
  • Adjust based on false positives (if you're getting too many)

Autonomy Modes ​

Control how much freedom your AI has:

  • Off: AI never sends automatically (draft-only mode)
  • Draft-Only: AI creates drafts but never sends them
  • Guarded Autonomy: AI can send, but with restrictions and review

Important: Hard stop conditions always override autonomy mode - even in guarded autonomy, hard stops will block sending. Safety first!

Safety Measures and Content Protection ​

The platform includes advanced safety measures to protect both your business and your customers:

Serious Mode Detection ​

The system automatically detects serious inquiries that require professional, empathetic responses:

What Triggers Serious Mode:

  • Legal threats or complaints
  • Safety concerns or injuries
  • Fraud or security issues
  • Discrimination or harassment
  • Billing disputes
  • Privacy violations
  • Medical emergencies
  • Extreme emotional distress

How It Works:

  • System automatically analyzes incoming messages
  • High-confidence patterns trigger serious mode immediately
  • Medium-confidence patterns are verified with AI classification
  • Serious mode activates for 30 minutes after detection
  • AI responses become more professional and empathetic in serious mode

Benefits:

  • Protects your business from legal issues
  • Ensures appropriate response to sensitive situations
  • Maintains professional tone during critical conversations
  • Automatic detection means no manual intervention needed

Content Safety Controls ​

The platform includes multiple layers of content safety:

PII (Personally Identifiable Information) Protection:

  • Sensitive information (SSN, credit cards, phone numbers, emails) is automatically redacted from AI context
  • PII is replaced with placeholders (e.g., , )
  • Placeholders are automatically replaced with actual values when sending responses
  • Protects customer privacy while maintaining conversation quality

Content Validation:

  • Responses are validated before sending
  • Profanity detection (configurable)
  • Off-topic content detection (configurable)
  • Maximum response length limits
  • OpenAI Moderation (optional): Model-based content moderation that detects harmful, inappropriate, or unsafe content using OpenAI's moderation models. Can be enabled for additional coverage beyond heuristic checks.

AI Safety Controls:

  • Temperature and token limits enforced
  • Model restrictions (allowed/blocked models)
  • Maximum function calls per response
  • Automatic clamping of unsafe configurations

Tool Use Improvements ​

The AI assistant can use built-in tools while maintaining safety:

Context Functions:

  • Product information lookup
  • Document search
  • Website content access
  • Calendar booking integration
  • Client information retrieval

PII Variable System:

  • Use and in responses
  • Variables are automatically replaced with customer's actual contact information
  • PII is never exposed to AI models directly
  • Maintains privacy while enabling personalization

Safety Features:

  • Limited function calls per response (prevents abuse)
  • Context functions respect tenant isolation
  • All tool usage is logged for audit
  • Automatic fallback if tools fail

Testing Your Configuration ​

After configuring your AI:

  1. Test with Demo Conversation: Create a demo conversation to see how it works
  2. Review Responses: Check if responses match your expectations
  3. Test Serious Mode: Try messages that should trigger serious mode detection
  4. Verify PII Protection: Check that sensitive information is handled correctly
  5. Adjust as Needed: Fine-tune prompts and settings based on results
  6. Monitor Performance: Track qualification rates and customer satisfaction

Migration to Assistants System ​

The platform has moved from a single AI configuration to a multi-assistant system. This provides:

  • Multiple Assistants: Create different assistants for different scenarios
  • Channel-Specific Defaults: Set default assistants per channel (email, SMS, chat, voice)
  • Routing Rules: Automatically select assistants based on conditions
  • Better Organization: Manage multiple AI personalities and configurations

To configure AI settings now:

  1. Go to AI Assistant → Assistants (/assistants)
  2. Create or edit assistants
  3. Set channel-specific defaults
  4. Configure routing rules if needed

Next Steps ​

autoch.at Documentation

Copyright © 2024 autoch.at