How an AI Agent works with an NLP bot
This section explains how AI Agent capabilities work alongside the existing NLP-based bot, how different settings influence query handling, and how to choose the right configuration based on your use case.
The platform provides two key GenAI settings:
- AI Agent (Bot Mandate)
- AI Intent Assist (Orchestrator)
These settings can be enabled independently or together to control how user queries are understood and answered. Navigate to these settings under Settings > Solution Settings; read more here: Solution Settings: Chatbot.
Core Concepts
NLP Bot (Default Behavior)
By default, bots use NLP intent detection based on:
- Keywords
- Training phrases
This works well for structured, predictable conversations, but may struggle with:
- Long or ambiguous queries
- Paraphrased questions
- Multiple intents in a single message
AI Agent
The AI Agent allows you to build GenAI-powered conversational flows using the AI Agent Studio.
Instead of relying on predefined intent matching, AI Agent flows:
- Use natural-language objectives and actions
- Handle open-ended and dynamic conversations
- Can retrieve knowledge (Exomind), call APIs, or perform logic-based actions
When enabled, AI Agent flows can generate responses directly using AI.
AI Intent Assist
AI Intent Assist improves how the bot understands user queries before deciding how to respond.
It introduces a Query Analyzer layer that:
- Understands the meaning of user queries
- Attempts to map queries to the closest NLP intent using AI
- The AI tries to read the user queries, all the existing NLP intents in the bot with their descriptions about what that intent does, and tries to match with the closest relevant intent.
- Reduces missed or incorrect intent detection
AI Intent Assist does not generate responses on its own; it only improves intent identification.
How Queries Are Processed
The behavior of the bot depends on which settings are enabled.
Scenario 1: Only NLP (Default)
Workflow:
- User sends a query
- NLP engine tries to match a predefined intent
- If matched → corresponding NLP flow is triggered
- If not matched → fallback or failure message is shown
Best for:
- Simple, structured bots
- Fixed menus and predictable queries
- It can integrate complex APIs
Scenario 2: Advanced NLP: NLP Workflow with AI Intent Assist On
Workflow:
- User sends a query
- NLP engine attempts keyword-based intent matching
- If no intent is found:
- It passes to the AI Intent Assist, where AI analyzes the query
- Attempts to identify the closest matching NLP intent (based on intent name and description)
- If intent is identified → Matched NLP flow is triggered
- If still not identified → failure or fallback message
Important Notes:
- AI is used only for intent detection, not response generation
- AI Agent flows are not used
Best for:
- Businesses that want better intent accuracy
- Customers who do not want AI-generated responses
- Improving NLP performance without changing bot behavior
Scenario 3: AI Agents
Workflow:
- User sends a query
- Query is routed directly to AI Agent flows
- AI Agent determines how to respond using objectives and actions
- Response is generated by AI
- No NLP intents are used.
Important Notes:
- A default AI Agent flow will be created, and the user will have to update the Prompts for the Supervisor & Sub-Goal nodes.
- NLP intent detection is not used
- The bot relies entirely on the AI Agent logic
- NLP flows are preserved but inactive
Best for:
- Fully GenAI-based bots
- Bots designed for free-flow conversations
Scenario 4: Hybrid Bot - NLP Workflow & AI Agent
This is the most complete and flexible setup.
Workflow:
- User sends a query
- NLP intent detection is attempted
- If no match:
- AI Intent Assist analyzes the query
- Attempts to map it to the closest NLP intent
- If an NLP intent is identified → NLP flow is triggered
- If no intent is identified:
- Query is routed to the AI Agent flows
- AI Agent generates the response and completes the flow.
- If the AI Agent cannot handle and generate a response → fallback message is shown
Key Benefits:
- Preserves structured NLP flows
- Uses AI only when needed
- Provides a safe fallback mechanism
- Reduces hallucinations and incorrect answers
Best for:
- Production bots
- Gradual GenAI adoption
- Combining structured flows with GenAI flexibility
- New gen bots
Current Limitations
- The platform currently supports NLP-first with AI Agent fallback
- AI Agent–first with NLP fallback is not supported
- Reason:
- NLP flows support richer UI components (CTAs, media, forms)
- AI Agent flows are still evolving toward feature parity
Once AI Agent flows support all NLP components, additional routing configurations will be introduced in a future release.
Choosing the Right Configuration
Use the table below to decide which settings to enable:
Use Case | Recommended Settings |
|---|---|
Only NLP Bot: Use only NLP Workflows (No Gen AI Features) - Default Option | AI Agents (Bot Mandate) = OFF AI Intent Assist (Orchestrator) = OFF |
Advanced NLP: Improve the intent accuracy of the NLP Bot, without AI responses | AI Agents (Bot Mandate) = OFF AI Intent Assist (Orchestrator) = ON |
Only AI Agents: Build a Gen AI-based Bot without any NLP Workflows | AI Agents (Bot Mandate) = ON AI Intent Assist (Orchestrator) = OFF |
Hybrid Bot: NLP Workflows followed by AI Agents | AI Agents (Bot Mandate) = ON AI Intent Assist (Orchestrator) = ON |
Best Practices
- Start with AI Intent Assist to improve NLP reliability
- Add an AI Agent gradually for fallback and complex queries
- Use AI Agent for:
- FAQs
- Knowledge-based queries
- Open-ended questions
- Keep transactional and UI-heavy flows in NLP until the AI Agent supports full parity