Agent Config Assistant

The Agent Config Assistant is an AI-powered conversational interface built into the Agent configuration dialog for designing and configuring your AI Agent through natural language. Describe what your Agent should do, iterate through conversation, and the Assistant will generate and refine the full configuration — use case, instructions, model, tools, capabilities, and introduction.

How it works

The Agent Config Assistant appears as a chat panel alongside the Agent configuration form. When you open the Agent Configuration dialog, you'll find the Assistant tab ready to help you build or refine your Agent.

Getting started

1. Open the Agent configuration dialog from the AI tab in Builder

2. Switch to the Assistant tab

3. Describe your Agent's purpose, target users, and goals

4. The Assistant generates and applies configuration changes in real time

The Assistant will ask clarifying questions to understand your requirements and then automatically populate the configuration fields — including the Use Case, Instructions, model selection, tools, capabilities, and introduction message.

What you can configure

The Assistant can help you set up every aspect of your Agent:

Configuration field	What the Assistant does
Use Case	Generates a concise mission statement based on your description
Instructions	Creates structured, comprehensive instructions following best practices
Model	Recommends and sets the appropriate LLM based on your complexity needs
Tools	Selects relevant MCP Tools from your available Workflows
Capabilities	Enables features like Query Sources when your use case requires data querying
Introduction	Crafts a welcome message and conversation starters tailored to your Agent's purpose

Iterative refinement

The Assistant supports multi-turn conversations, so you can refine the configuration incrementally:

• Ask the Assistant to adjust specific sections of the Instructions

• Request changes to the model or tools selection

• Add new scenarios or edge cases to the Agent's behavior

• Update the introduction message or conversation starters

Each time the Assistant makes changes, it provides a rationale explaining why each update was made, helping you understand the reasoning behind the configuration choices.

The Assistant is context-aware: it knows your map's datasets, layers, widgets, and available tools. This means its recommendations are grounded in your actual data and map configuration.

Combining with manual edits

The Assistant and the manual configuration form work together seamlessly:

• Changes made by the Assistant are immediately reflected in the form fields

• Manual edits you make in the form are picked up by the Assistant in subsequent messages

• You can switch between the Assistant and form tabs at any time without losing progress

Changing the model in the Assistant will reset the conversation history, as different models may produce different configuration styles. You'll be asked to confirm before this happens.

Best practices

• Describe the problem, not the configuration: Focus on what users need to accomplish and who they are. The Assistant already knows your map's datasets, layers, widgets, and available tools — it will reference them automatically in the generated configuration.

• Test and refine mid-conversation: Run your Agent at any point during configuration to see how it behaves, then come back to the Assistant with specific feedback. The Assistant will refine surgically — preserving what works and adjusting only what you flag.

• Review the rationale: Each configuration update includes a separate explanation of why each change was made. Use this to verify the Assistant's reasoning aligns with your intent.

• Let the Assistant ask questions: The Assistant may ask clarifying questions before generating configuration — such as who will use the Agent, what questions they should be able to ask, or whether custom SQL queries are needed. This produces better results than providing all details upfront.