Platform Mental Model

This page explains the major building blocks of Airgentic and how they work together. Understanding this mental model will help you operate the platform effectively and make sound decisions about configuration.

The big picture

Airgentic is an AI-powered platform that helps your customers get answers, complete tasks, and connect with your team when needed. It does this by combining:

1. Knowledge — Your website content and uploaded documents, indexed and searchable
2. AI understanding — Language models that understand questions and generate helpful responses
3. Agents — Specialist AI personalities that handle different types of enquiries
4. Actions — Custom functions that can search, send emails, call APIs, and more
5. Human oversight — Tools for monitoring, intervening, and maintaining quality

These components work together to deliver a complete customer service experience.

Core components

Knowledge sources

Everything Airgentic knows comes from your knowledge sources:

The platform extracts text, processes it, and stores it in a searchable index. When users ask questions, the AI searches this index to find relevant information.

Related documentation:
- Data Sync — Running and scheduling crawls
- Crawler Settings — Controlling what gets indexed
- Upload Documents — Adding files directly

Search

Search is the foundation of how the AI finds information to answer questions. When a user asks something:

1. The AI formulates a search query based on the question
2. The search engine retrieves relevant pages from your index
3. Results are ranked and filtered according to your search settings
4. The most relevant content is passed to the AI to formulate a response

You can configure how results are categorised, adjust ranking with boosts, and add synonym rules to improve matching.

Related documentation:
- Search Settings — Categories, mappings, and synonyms

AI answers

When the AI has search results, it generates an answer by:

1. Reading the relevant content from your knowledge sources
2. Understanding the user's question in context
3. Composing a natural language response
4. Citing the sources it used

The AI is designed to stay grounded in your content — it should not make things up. If it cannot find relevant information, it tells the user it cannot answer.

Agents

Agents are specialist AI assistants that handle different types of questions. Every service has:

• Frontline agent — The default agent that handles all questions and decides whether to delegate to a specialist
• Specialist agents — Optional agents you create for specific domains (e.g., technical support, sales enquiries, product lookup)

Each agent has its own:
- Role description — What types of questions it handles (used by Frontline to decide when to delegate)
- Prompt — Instructions that define its personality, tone, and behaviour
- Functions — Actions it can perform (search, send email, call APIs)
- Search scope — Which parts of your content it can access

When a user asks a question, the Frontline agent reads the question and decides whether to handle it directly or delegate to a specialist.

Related documentation:
- Agents Overview — Managing agents
- Creating an Agent — Step-by-step setup
- Editing an Agent — Configuration options

Prompts

Prompts are the instructions that tell agents how to behave. A prompt typically includes:

• Identity — Who the agent is and what organisation it represents
• Tone — How it should communicate (formal, friendly, concise, etc.)
• Guardrails — What it should and should not do
• Output format — How to structure responses

Airgentic uses modular prompts — reusable components that can be shared across agents. This keeps instructions consistent without duplication.

Related documentation:
- Edit Prompts — Managing prompts
- Editing a Prompt — Writing and version control

Functions

Functions let agents perform actions beyond answering from the knowledge base:

Functions are written in AirScript (a secure subset of Python) and defined with schemas that tell the AI when and how to use them.

Note: Functions are an advanced capability. Many services work well using only the built-in search and answer functionality.

Related documentation:
- Functions Overview — What functions are and how they work
- AirScript Functions — Technical reference

Curated answers

Curated answers let you override the AI with exact, approved content for specific questions:

• Verbatim answers — The exact text (or HTML) the agent will show
• Pinned URLs — A specific page the AI must use for that question

Use curated answers when:
- The answer isn't in your content
- Multiple sources conflict and you want one authoritative answer
- Exact wording is required (legal, policy, contact details)

Related documentation:
- Curated Answers — Setup and best practices

Human handoff

Human handoff lets your team monitor conversations and intervene when needed:

• AI Only mode — AI handles everything
• Human-Supported AI mode — AI responds, but humans can monitor and take over
• Human Only mode — AI is disabled; humans respond to everything

The Human Handoff screen shows active conversations in real time, with automatic flags for conversations that need attention (thumbs down, failed answers, negative sentiment, help requests).

Related documentation:
- Human Handoff — Monitoring and intervention

Secure services

Secure services restrict access to authenticated users only — ideal for internal use cases like staff intranets. With secure services:

• Users sign in through your organisation's identity provider (Microsoft Entra ID, Okta, Google Workspace)
• Only authorised users (by email domain, individual email, or group membership) can use the service
• No passwords are entered into the Airgentic widget

Related documentation:
- Secure Services Overview — Setup and authorisation options
- How Secure Authentication Works — Technical explanation

Voice mode

When voice mode is enabled, users can speak to the AI using their microphone. The AI responds with both text and audio. Voice mode uses the same agents and knowledge sources as text conversations.

Related documentation:
- Voice Mode Settings — Configuration options

Insights and analytics

Insights help you understand how your service is performing:

Use insights to identify content gaps, spot problem areas, and track improvement over time.

Related documentation:
- Customer Service Insights — Conversation analytics
- Search Insights — Search analytics

How the components work together

Here's a simplified view of what happens when a user asks a question:

1. User sends a message via the chat widget (or speaks via voice mode)
2. Frontline agent receives it and decides whether to handle it or delegate to a specialist
3. Selected agent processes the question using its prompt and available functions
4. Search retrieves relevant content from your indexed knowledge sources
5. AI generates a response grounded in the search results
6. Response is shown to the user with citations to sources
7. Conversation is logged for insights and (optionally) human review

If curated answers exist for the question's intent, they take priority. If the agent has functions enabled, it may call them (e.g., to look up additional information or escalate to a human).

What this means for operators

Understanding this model helps you:

• Know where to look when something isn't working (content? agent routing? prompt? search settings?)
• Make informed decisions about what to configure vs what to leave alone
• Explain the system to colleagues and stakeholders
• Plan improvements systematically rather than guessing

Related pages

• What to Use When — Choosing the right capability
• Admin and Operator Essentials — Practical tasks
• Deep-Dive: Agents and Prompts — More detail on agents
• Deep-Dive: Search and Knowledge Sources — More detail on search

Next: Admin and Operator Essentials