Agenteditor

Agent Editor Workflow Guide

Quick reference for using Agent Editor in Studio Pro 11.12.0+ to define, test, and deploy GenAI agents.

See GLOSSARY for detailed term definitions and technical details.

Document Types

Agent Editor provides four document types (add via Add other menu at module level):

• Model: Links to Mendix Cloud GenAI text generation resource
• Agent: Defines prompts, tools, knowledge bases, and settings
• Knowledge Base: Configures RAG (retrieval-augmented generation)
• Consumed MCP Service: Connects to external MCP tool servers

Configuration Workflow

1. Define Model (Required)

• Create Model document
• Set constant with GenAI resource key
• Click "List Models" to validate

2. Define Agent (Required)

• Create Agent document, select Model
• Write System prompt (use {{placeholders}} for dynamic values)
• Select Context entity if using placeholders
• Optional: Configure model settings (max tokens, temperature, TopP)

3. Add Tools (Optional)

• Microflow tools: Select microflow returning String, add Name/Description
• MCP tools: Create Consumed MCP Service first, configure endpoint/credentials
• Use Active checkbox to enable/disable during testing

4. Add Knowledge Bases (Optional)

• Create Knowledge Base document
• Link to agent, select collection
• Add Name/Description for LLM to understand when to use it
• Configure Max results and Min similarity threshold

5. Test

• Switch to Playground mode (top right switcher)
• Requires: app running locally, no errors, after-startup configured
• Click Test button, review Console logs
• Iterate: edit in Build mode → restart app → test in Playground

6. Integrate

For Microflows:

• Use "Call Agent without History" for task agents (single call)
• Use "Call Agent with History" for chat agents (multiple messages)
• Pass context object when using placeholders

For Conversational UI:

• Use "New Chat for Agent" to create a ChatContext for the selected Agent
• Pass chat context to conversational UI snippets or pages
• Enables multi-turn conversations with history

7. Deploy

• Agent documents deploy with app model
• Override constants per environment (keys, endpoints)

Prompts and Placeholders

System Prompt (Required)

Defines agent behavior, personality, and capabilities.

User Prompt (Optional)

Template for task-style execution with placeholders.

Placeholders

Use {{attributeName}} syntax - must match Context entity attribute names exactly.

Example:

System: "You are a {{role}} assistant for {{companyName}}."
Context Entity: ServiceContext with attributes: role, companyName

Agent Modes

Switch between modes in top right corner of Agent Editor:

Build Mode: Configure prompts, tools, knowledge bases, settings

Playground Mode: Test interactively

• Requires running app
• Test one shot task agents or multi-turn chat agents
• See real-time tool calls and responses

Common Patterns

Task Agent: System + User prompts, optional multiple tools and KBs Chat Agent: System prompt only, optional multiple tools and KBs

Troubleshooting Quick Reference

First-Time Setup

After installing Agent Editor from Marketplace:

1. Add /agenteditor to .gitignore (not excluded automatically)
2. Set 32-character encryption key (App > Settings > Configuration)
3. Set ASU_AgentEditor as after-startup microflow (App > Settings > Runtime)
4. Install all required dependencies
   • View compatible versions: Extensions menu > Agent Editor > Compatibility
   • Opens module dependency overview showing required modules and compatible versions
5. Obtain GenAI resource keys from Mendix Cloud GenAI Portal
6. Create constants for keys

Module Usage Rules

CRITICAL: Agent Editor Commons is Read-Only

What You CAN Do

• Reuse logic from the USE_ME folder: Copy patterns, reference examples, and use template agents from the USE_ME folder
• Call the documented toolbox actions: Use "Call Agent without History" and "Call Agent with History" actions in your microflows
• Use the documented domain model entities: Work with Response, Request, and other GenAI Commons entities
• Create Agent documents: Define your own agents using the Agent Editor extension

What You CANNOT Do

• Modify Agent Editor Commons logic: Do NOT change any microflows, pages, or logic within the Agent Editor Commons module
• Use private folder logic: Do NOT call or reference any microflows, entities, or logic from private folders
• Extend the module directly: Do NOT add custom logic to the Agent Editor Commons module

Extension Pattern

If you need to extend or customize Agent Editor functionality:

1. Create a separate module in your app (e.g., MyCustomAgents)
2. Import or reference only the documented toolbox actions and domain model from Agent Editor Commons
3. Implement your custom logic in your own module
4. Use agents and microflows from the USE_ME folder as templates, but implement them in your module

This separation ensures:

• Your customizations survive module updates
• Module upgrades don't break your custom logic
• Clear separation between platform code and application code
• Compliance with Mendix module development best practices

Limitations

• Only Mendix Cloud GenAI supported (more providers planned)
• Mac support limited
• MCP whole-server only (no individual tool selection yet)
• Tool Choice "Tool" only for microflows currently

Reference Links

Access helpful resources via Extensions menu > Agent Editor > Onboarding in Studio Pro.

• Agent Editor Docs
• Detailed Glossary