Airgentic Help
Airgentic offers several capabilities for different needs. This page provides a practical decision framework for choosing the right approach.
| I want to... | Use this |
|---|---|
| Answer questions from my website content | Default Frontline agent + search |
| Ensure exact wording for specific answers | Curated Answers (verbatim) |
| Force the AI to use a specific page | Curated Answers (pinned URL) |
| Handle different question types differently | Specialist agents |
| Perform actions (email, API calls, lookups) | Functions |
| Restrict access to authenticated users | Secure Services |
| Let users speak instead of type | Voice mode |
| Monitor conversations and intervene | Human Handoff |
| Understand how the service is performing | Insights |
Use it when:
- Questions can be answered from your website content or uploaded documents
- You want the AI to synthesise information from multiple pages
- You don't need exact control over wording
- The standard Frontline agent behaviour is appropriate
How it works:
The Frontline agent receives the question, searches your indexed content, and generates a response based on what it finds. Citations link to the source pages.
This is the baseline. Most questions should be handled this way. Only add more complex configuration when there's a specific reason.
Use it when:
- The answer isn't available in your indexed content (opening hours, specific contact details)
- Multiple sources conflict and you need one authoritative answer
- Exact wording is required (legal disclaimers, policy statements)
- You want to deflect certain topics in a specific way
- You want to guarantee a particular page is cited
Two modes:
| Mode | Use when |
|---|---|
| Verbatim answer | You want to control the exact response text |
| Pinned URL | You want the AI to always use a specific page for that topic |
How to implement:
Add the curated answer via Edit Curated Answers. The AI matches by intent (not exact wording), so write a representative question.
Related: Curated Answers
Use it when:
- Different question types need different handling (e.g., sales vs support vs technical)
- You want different prompts, tones, or behaviours for different domains
- You want to restrict search scope for certain topics (e.g., product support agent only sees product pages)
- You want to enable specific functions for specific question types
How it works:
When a question comes in, the Frontline agent evaluates it against the role descriptions of all specialist agents. If a specialist is a better fit, the Frontline delegates to it.
Be thoughtful about adding agents. Each agent adds complexity. Start with the Frontline and only add specialists when there's clear differentiation.
Related: Agents Overview, Creating an Agent
Use it when:
- You need to perform actions beyond answering questions
- You want to search with specific filters or scopes
- You need to call external APIs (order lookup, CRM, booking systems)
- You want to send emails (enquiry escalation, lead capture)
- You want to display maps or other rich content
Examples:
- Escalate to human — collect customer details and email to support
- Store finder — geocode location and display map markers
- Order status — look up order details from e-commerce API
- Product filter — search only pages matching a product model
Be aware:
Functions require AirScript code and JSON schemas. Global Functions are provided for common use cases. Custom functions need development effort.
Related: Functions Overview, AirScript Functions
Use it when:
- The service is for internal users only (staff intranet, internal knowledge base)
- Content contains sensitive or confidential information
- You need to restrict access by email domain, group membership, or specific users
How it works:
Users sign in through your organisation's identity provider (Microsoft Entra ID, Okta, Google Workspace). Only authorised users can access the chat and search.
Be aware:
Secure services require coordination between your IT team and Airgentic to configure the identity provider integration.
Related: Secure Services Overview
Use it when:
- Users may prefer to speak rather than type
- You're targeting accessibility needs
- The use case benefits from conversational voice interaction
How it works:
When enabled, a microphone button appears in the widget. Users can speak their question, and the AI responds with both text and audio.
Be aware:
Voice mode uses the same agents and knowledge as text. Voice responses may be phrased slightly differently for spoken delivery.
Related: Voice Mode Settings
Use it when:
- You want to monitor conversations in real time
- You need the ability to intervene when the AI struggles
- You offer live support as an escalation path
- You want to catch and respond to urgent issues quickly
Three modes:
| Mode | When to use |
|---|---|
| AI Only | No human monitoring needed; AI handles everything |
| Human-Supported AI | AI responds normally, but humans can watch and take over when needed |
| Human Only | AI is fully disabled; humans respond to everything (rare) |
Related: Human Handoff
Use it for:
- Understanding how the service is performing overall
- Identifying questions the AI struggles to answer
- Reviewing flagged conversations (thumbs down, negative sentiment, failed answers)
- Finding content gaps
- Tracking trends over time
- Reporting to stakeholders
Key views:
| Tab | What it shows |
|---|---|
| Insights | Summary stats, charts, top citations, products mentioned |
| Threads | Individual conversations with full transcripts |
| Geography | Where users are located |
| Timelines | Trends over time |
Related: Customer Service Insights, Search Insights
Start here: Can the question be answered from your content?
│
├── YES → Is exact wording required?
│ ├── YES → Use Curated Answers (verbatim)
│ └── NO → Use default search and answer
│
├── NO → Is the answer available on a specific page?
│ ├── YES → Use Curated Answers (pinned URL)
│ └── NO → Do you need to perform an action (email, API call)?
│ ├── YES → Use Functions
│ └── NO → Add the information to your content and re-sync
│
└── Does this question type need different handling?
├── YES → Consider a Specialist Agent
└── NO → Use the Frontline agent
These capabilities work together:
Start simple. Add capabilities as needs emerge.
| Scenario | Recommended approach |
|---|---|
| Public website FAQ bot | Frontline agent + curated answers for key questions |
| Internal HR policy assistant | Secure service + uploaded documents + curated answers for common policies |
| Product support with order lookup | Specialist agent + function for order status API |
| Multi-department with different tones | Multiple specialist agents with distinct prompts |
| Live-supported service | Human-Supported AI mode + Human Handoff monitoring |