Skip to main content

​
Prerequisites

Before starting, make sure you have sample data loaded so the AI assistant has entities to work with.
This guide assumes you have completed the Quickstart and loaded the sample dataset. If you haven’t yet, follow the Sample Data guide first.

​
Open the AI Panel

Click the sparkle icon in the top navigation bar to open the AI assistant panel, or use the keyboard shortcut:
  • Mac: Cmd + I
  • Windows / Linux: Ctrl + I
The panel slides open on the right side of the screen. You can type natural language queries and the assistant will route them to the appropriate tools automatically.

​
Try These Conversations

Work through these five examples to see what the AI assistant can do.
1

Ask about your offers

Type:
How many offers do I have?
The AI calls the listOffers tool, returns the total count, and lists each offer by name with its category and status. This is a read operation, so it executes immediately with no confirmation needed.
2

Check scoring models

Type:
What scoring models are configured?
The AI calls listModels and returns each model’s name, type (propensity, uplift, rule-based), and performance metrics like AUC. Use this to quickly audit your model inventory.
3

Detect policy conflicts

Type:
Are there any contact policy conflicts?
The AI calls analyzePolicyConflicts to scan your contact policies for contradictions — for example, one rule suppressing a channel while another requires it. The response highlights any conflicts with severity and suggested fixes.
4

Create a new channel

Type:
Create a new channel called SMS with type sms and delivery mode api
This is a mutation, so the AI shows a preview of the channel it will create and asks you to approve or reject. Click Approve to proceed. The assistant then calls createChannel and confirms the result.
All write operations (create, update, delete) follow this guided autonomy pattern — you always see what will change before it happens.
5

Run a recommendation

Type:
Run a recommendation for customer cust-001 with age 30 and income 75000
The AI calls the recommend tool with the customer ID and attributes you provided. It returns the ranked list of eligible offers, each with its score, creative, and channel assignment. This is a great way to test your Decision Flows without writing API calls.

​
How It Works

The AI assistant is powered by a tool-use architecture that connects natural language to platform operations.
  • 130+ MCP-aligned tools organized by category: studio, data, algorithms, intelligence, and docs. Every entity in the platform can be created, read, updated, or deleted through conversation.
  • Guided autonomy: Read operations execute immediately and return results. Mutations (create, update, delete) show a preview with an approve/reject prompt so you stay in control.
  • Context-aware routing: The AI understands your data model and routes queries to the right tool automatically. Ask “show me my pipelines” and it calls listPipelines; ask “explain this flow” and it calls getDecisionFlow.
  • Multi-provider support: The assistant supports Google Gemini (default), Anthropic Claude, OpenAI, Amazon Bedrock, and Ollama for local models. Configure your preferred provider in Settings > AI Configuration.

​
What’s Next

AI Assistant Docs

Full reference for all assistant capabilities, tool categories, and conversation patterns.

AI Configuration

Set your preferred AI provider, API keys, and model parameters.

MCP Integration

Connect KaireonAI tools to external AI IDEs like Cursor, Windsurf, and Claude Code.