Core Module
AI Assistants
AI Assistants (AI Bots) answer calls using conversational AI, play tailored prompts, and transfer callers when necessary. Each assistant bundles speech models, voice prompts, and fallback routing so you can automate common requests without sacrificing customer experience.
Module Workflow
- Visit AI Assistants in the navigation.
- The table lists every bot with quick links to call logs, conversations, and edit/delete actions.
- Click Create AI Assistant (or edit) to configure voices, AI providers, and routing.
Configuration Steps
- Identity & Voices - Name the assistant and assign the welcome voice plus optional tones (waiting, inaudible, listening, transfer). Built-in preview controls let you verify recordings instantly.
- AI Provider Profiles - Select the LLM provider profile (required) that powers conversation logic, along with optional TTS (text-to-speech) and STT (speech-to-text) profiles if you want to override defaults.
- Assistant Instructions - Use the rich textarea to describe the bot's purpose, personality, escalation rules, and allowed actions. This prompt becomes the system message guiding every interaction.
- Behavior Controls - Configure max interactions, silence thresholds, notification email, and whether unresolved conversations should automatically create support tickets.
- Fallback Destination - Choose the module and destination that should take over if the AI cannot assist (e.g., queue, agent, voicemail).
- Save the assistant. The UI automatically re-populates destination lists via AJAX when you change modules.
Operational Tips
- Clear Prompts - Spell out what the bot should do, information it may request, and when to escalate. Include tone guidelines ("friendly, concise") to keep responses on-brand.
- Voice Consistency - Pick voice files that match your company persona; consider uploading custom audio for premium experiences.
- Monitoring - Review conversation transcripts (conversation.blade.php) and AI-assisted call logs to improve prompts or escalate knowledge gaps.
- Safety Nets - Keep Max Interactions and Max Silence conservative so callers are not trapped in loops.
- Notifications - Use the email field and optional ticket creation to alert humans whenever the AI transfers or fails to understand a request.
Field Reference
| Section | Field | Required | Description |
|---|---|---|---|
| Identity | Name | Yes | Display name shown throughout the UI. |
| Voices | Welcome Voice | Yes | Plays when the call begins; other tones are optional but improve polish. |
| AI Profiles | LLM Provider Profile | Yes | Selects which AI backend handles reasoning. |
| AI Profiles | TTS Profile, STT Profile | No | Overrides the default speech stack if necessary. |
| Instructions | Assistant Instructions | Yes | System prompt describing goals, allowed data, escalation triggers, etc. |
| Behavior | Max Interactions, Max Silence, Email, Create Support Ticket | No | Guardrails to prevent overly long or stalled conversations; optionally send transcripts to email or open tickets. |
| Failover | Fallback Destination Module, Destination | Yes | Where calls route when the AI transfers/escalates. |
*Destination select refreshes automatically when you change the module via function_id.
Troubleshooting
- Destination Dropdown Empty: Ensure the module dropdown has a value; the destination list relies on AJAX calls to populate.
- Voices Not Playing: Confirm the uploaded files exist and that browser autoplay policies allow preview playback.
- AI Seems Confused: Refine the Assistant Instructions with concrete steps, example phrases, and escalation criteria. Also verify the correct LLM profile is selected.
- Support Tickets Not Creating: Check the checkbox state and make sure the ticket module is enabled for your user role.
- AI Assistants thrive with iteration. Review caller feedback often, update prompts, and adjust fallback logic as your automation maturity grows.
Practical Example
Scenario: Deploy a concierge bot that greets callers, answers FAQs, and transfers to the live reception queue when it detects frustration.
- Name: "Concierge Bot"
- Welcome Voice: concierge_intro.wav
- LLM Provider: gpt4o-enterprise
- Instructions: "You are a warm receptionist. Answer billing/shipping/status questions. When the caller says 'agent', 'representative', or expresses frustration, transfer to Reception Queue."
- Max Interactions: 12; Max Silence: 4
- Email: concierge-alerts@company.com
- Create Support Ticket: enabled (captures transcripts when the bot escalates)
- Fallback Module: queue; Destination: Reception Queue
Maintenance & Automation
Routine Checks
- Review transcripts weekly, tagging misunderstandings or escalation triggers that need prompt tweaks.
- Rotate LLM API keys or provider credentials in alignment with your security policy.
- Confirm fallback destinations remain active whenever you reorganize queues or IVRs.
Related Platform Pages
Connect this module to product and solution context for planning and implementation.