Agent Mode Overview

APD includes an AI-powered Agent Mode that helps you design and build portals through natural language conversation. The agent can create actions, templates, and content by understanding your requirements and using built-in tools.

What is Agent Mode?

Agent Mode is a conversational AI assistant that:

Understands portal design requirements
Creates and modifies actions, templates, and content
Queries Business Central tables and fields
Compiles actions automatically
Maintains context across sessions

Supported AI Providers

APD supports multiple AI providers:

Provider	Description	Best For
Azure OpenAI	Microsoft Azure-hosted OpenAI	Enterprise with Azure subscription
ChatGPT OpenAI	Direct OpenAI API	General use
Anthropic (Claude)	Anthropic's Claude models	Complex reasoning tasks
LM Studio	Local AI model hosting	Offline/privacy requirements

Agent Capabilities

The agent has access to these tools:

Action Management

Tool	Description
`list_actions`	List all existing actions
`get_action`	Get details of a specific action
`new_action`	Create a new action
`update_action`	Update a field on an action
`compile_action`	Compile a specific action
`compile_all`	Compile all actions

Template Management

Tool	Description
`list_templates`	List all templates
`search_templates`	Search templates by name and HTML body
`get_template`	Get HTML content of a template
`save_template`	Save HTML content to a template

Content Management

Tool	Description
`list_content`	List all content items
`get_content`	Get content (markdown)
`save_content`	Save content

Script Management

Tool	Description
`get_script`	Get the script source of a Script-type action
`save_script`	Save and compile the script of a Script-type action

Security

Tool	Description
`add_profile`	Create a security profile (used by the `<?profile?>` tag and on action security settings)

Business Central Discovery

Tool	Description
`get_tables`	Search for BC tables by name
`get_fields`	Get all fields in a specific table

How Agent Mode Works

┌─────────────────────────────────────────────────┐
│                 Your Request                     │
│   "Create a customer list page with name,       │
│    city, and balance columns"                   │
└─────────────────────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────┐
│               AI Processing                      │
│   Agent understands requirements and plans      │
│   tool calls                                    │
└─────────────────────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────┐
│               Tool Execution                     │
│   1. get_tables('Customer')                     │
│   2. get_fields(18)                             │
│   3. new_action('CUSTOMER-LIST')                │
│   4. save_template(...)                         │
│   5. update_action(..., 'HTML Template')        │
│   6. compile_action('CUSTOMER-LIST')            │
└─────────────────────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────┐
│                   Result                         │
│   Action and template created and compiled      │
└─────────────────────────────────────────────────┘

Agent Knowledge

The agent understands:

APD template tag syntax (<?field()?>, <?loop()?>, etc.)
Navigation between actions (<?action()?>)
Template composition with <?parse()?> includes
Expression syntax and functions
Table and field naming conventions

Session Memory

The agent maintains notes across sessions through a special content item ai.progress. This allows the agent to:

Remember project context
Track what has been built
Maintain consistency in naming

Custom Instructions

You can provide project-specific instructions by creating a content item called ai.instructions. The agent will include these in its context for every conversation.

Example Instructions

# Portal Instructions

This is a customer self-service portal.

## Design Guidelines
- Use Bootstrap 5 for styling
- All pages should include the header and footer templates
- Customer data must be filtered to the logged-in customer

## Naming Conventions
- Actions: Use KEBAB-CASE (e.g., CUSTOMER-LIST)
- Templates: Use lowercase with hyphens (e.g., customer-list)

Getting Started

Configure Agent Setup
Open Agent Development page
Start conversing with the agent
Review and test generated components