Automation AI - Kore.ai Docs

Build conversational AI assistants for customers, agents, and employees across voice and digital channels. Automation AI uses DialogGPT and AI Agents to deliver multi-turn conversations that understand intent, retain context, and respond naturally.

Key Components

┌──────────────────────────────────────────────────────┐ │ Automation AI │ └───────────────────────────┬──────────────────────────┘ │ │ │ ┌────────────────┼────────────────┐ │ ▼ ▼ ▼ │ ┌─────────────┐ ┌─────────────┐ ┌──────────────┐ │ │ DialogGPT │ │Agent Flows │ │ AI Agents │ │ │─────────────│ │─────────────│ │──────────────│ │ │ Agentic │ │ Dialog Task │ │ Tool-Calling │ │ │orchestration│ │ + │ │ + │ │ │ No training│ │ Agent Nodes │ │ External │ │ │ data needed│ │ │ │ Integrations │ │ └──────┬──────┘ └──────┬──────┘ └──────┬───────┘ │ └────────────────┼────────────────┘ │ │ │ ┌────────────────┴────────────────┐ │ ▼ ▼ │ ┌──────────────────┐ ┌──────────────────┐ │ │ Conversation │ │ Evaluation │ │ │ Management │ │──────────────────│ │ │──────────────────│ │ Testing Suite │ │ │ Interruptions │ │ Validate flows │ │ │ Clarifications │ │ Pre-deployment │ │ │ Context Switches │ │ checks │ │ └──────────────────┘ └──────────────────┘

Component	Description
DialogGPT	Agentic orchestration engine that routes conversations using generative models—no training data required.
Agent Flows	Conversational workflows combining and Agent Nodes for goal-driven service interactions.
AI Agents	Agent Nodes with tool-calling that handle complex tasks via contextual intelligence and external integrations.
Conversation Management	Handles interruptions, clarifications, and context switches mid-conversation.
Evaluation	Testing suite to validate conversational workflows before deployment.

DialogGPT Orchestration

DialogGPT analyzes each user message and routes it to the appropriate handler:

┌─────────────────────────────────────────────────────────┐ │ User Message │ └──────────────────────────┬──────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────┐ │ DialogGPT Orchestrator │ │ │ │ Analyzes intent, context, and confidence to route to: │ │ │ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ │ │ Agent │ │ Generative │ │ Agent │ │ │ │ Flows │ │ AI │ │ Handoff │ │ │ │ │ │ │ │ │ │ │ │ Structured │ │ LLM-powered │ │ Transfer │ │ │ │ tasks │ │ │ │ to human │ │ │ └─────────────┘ └─────────────┘ └─────────────┘ │ └─────────────────────────────────────────────────────────┘

Build an AI Agent

Follow these steps to create, configure, test, and deploy an AI Agent.

1. Create and Configure the App

Step	What to do
Create the app	Create a DialogGPT-based app. The platform auto-enables the required XO GPT models; Dialogs and FAQs are on by default. Guided Onboarding →
Configure DialogGPT	Define Conversation Types and set models for Chunk Shortlisting and Conversation Orchestration. Enable or disable Intent and Conversation Events and override defaults as needed. Conversation Orchestration →

2. Configure Generative AI

Step	What to do
Integrate an LLM	Connect to a supported LLM provider, a bring-your-own model, or Kore.ai XO GPT. For Tool Calling, Streaming Responses, or Dynamic Variables, use a custom prompt with a pre-built or custom integration. LLM Integration →
Create a custom prompt	Tailor model behavior per use case—build from scratch or import an existing prompt. Agent Node supports tool calling and prompt streaming with OpenAI/Azure OpenAI response formats in custom JavaScript V2 prompts. Prompts Library →
Enable GenAI features	Activate LLM-powered features that accelerate development and improve runtime performance. Features must be explicitly enabled before use. GenAI Features →
Configure data safeguards	Enable PII/sensitive data anonymization at the assistant and LLM level. Data Anonymization → Set up Guardrails to enforce appropriate AI outputs. Guardrails →

3. Build Flows and Connect Knowledge

Step	What to do
Create a Dialog	Define conversation flows using interlinked nodes. Nodes retrieve data, perform actions, call external apps, send messages, and control branching logic. Dialog Tasks →
Add an Agent Node	Use LLMs and tool calling to handle complex tasks, collect entities, and integrate with external systems. Supports multilingual conversations and contextual intelligence. Agent Node →
Connect Search AI	Index content from websites, documents (PDF, Office), and third-party systems (ServiceNow, Confluence, etc.) to give DialogGPT a reliable knowledge base. Content Sources →
Add supporting nodes	Use Prompt, Entity, and Agent Transfer nodes with transitions to complete the conversation flow. Nodes Overview →

4. Test

Step	What to do
Interactive testing	Validate your app in real time using the built-in Playground before publishing. Playground →
Batch testing	Upload CSV or JSON test cases to validate accuracy and reliability at scale. Generates comprehensive performance metrics. Batch Testing →

5. Deploy

Step	What to do
Enable a channel	Connect the agent to one or more voice or digital channels. The agent isn’t accessible to users until at least one channel is enabled. Digital Channels →
Configure agent transfer	Set up to a human agent. Integrations are platform-hosted—no custom BotKit required. Agent Transfer →
Publish	Submit the app through the publishing flow for admin review before making it available to end users. Publishing →

​Key Components

​DialogGPT Orchestration

​Build an AI Agent

​1. Create and Configure the App

​2. Configure Generative AI

​3. Build Flows and Connect Knowledge

​4. Test

​5. Deploy