Custom AI agents

Sitefinity CMS provides a new AI agents framework that enables you to define and create custom AI agents in your Sitefinity backend. You can use your custom AI agents in parallel with the built-in AI agents or independently. The AI agents framework allows custom agents to be configured through system prompts and settings individually, as described below in this article.

Upon interacting with the custom agent, content editors can apply consistent personalized recommendations for Stefinity CMS content, which are displayed in-context during their authoring experience.

If you want to create and configure a custom AI agent in your Sitefinity CMS backend, you can do so via the AI agents module.
Perform the following:

1. Log in to your Sitefinity CMS backend with Administrator privileges.
2. Navigate to Administration » AI agents and click Create AI agent on the AI agents screen.
3. In Agent name, enter the name of the custom agent.
   For example, enter Clarity & Readability.
4. In Enable this agent, you can optionally disable the AI agent by selecting No.
   The default setting is Yes.
5. In Instructions, you define a system prompt as per your individual needs. Manually feed your guidelines (no specific format is required) to instruct the AI agent. These instructions should guide how the AI agent analyzes the content and performs its tasks.
   
6. For multisite setup, under This agent works with..., select the desired option:
   • All sites (default option) - use this option if you want your AI agent to work consistently across all your sites.
   • Selected sites - use this option if you want your AI agent to work only for the site(s) you select here.
7. Under For content types and fields..., select the desired option:
   • All content types and fields - use this option if you want your AI agent to suggest improvements for all available content types and fields.
   • Selected content types and fields... - use this option if you want your AI agent to suggest improvements only for content types and fields that you specify here.
     The agent will analyze all fields of the selected content types, but suggestions will be generated only for the content in the selected/added fields. For example, a Clarity & Readability agent may be configured to suggest improvement only for Description and Content fields of News content type.
8. In the Agent identity section:
   • In Short name, enter a shorter version of your AI agent name.
     The short name is used where space is limited in the UI. For example, you can enter Readability.
     If short name is not provided, the Agent name value will be used.
   • In Agent ID, you can provide an unique identifier for your AI agent (used internally by the AI agents workflow orchestration).
     For example, enter clarity-and-readability.
9. Click Create.
   

Once created and configured, your custom AI agent is integrated directly into the Sitefinity content editing experience and is ready to surface AI-powered suggestions for existing or new content items.

When opening an existing content item for editing or creating a new one (such as Blogs, News, and Events), the AI agent analyzes the affected fields and generates different recommendations - according to the configured behavior of the agent.
When multiple AI agents are activated for the same task (including the built-in AI agents), all agents are executed in a strictly coordinated order and some of the generated AI suggestions may overlap - these suggestions are already aggregated and merged before being surfaced.  
Content editors get a notification icon (Content suggestions) on their active screen indicating the counter of all AI-generated suggestions for the currently open content item.

By clicking the notification icon, content editors access all AI-generated suggestions integrated in the Content suggestions sidebar on the screen.
The provided suggestions are actionable and are grouped by the individual sections and fields of the currently open content item. 

User actions on AI-generated suggestions

NOTE: No content changes can be applied by the AI agent itself without a human in the loop.

When reviewing the AI-generated suggestions in detail, content editors can perform the following actions available in the Content suggestions sidebar:

• Filter the AI suggestions available on the user interface, provided multiple AI agents are configured and enabled in your Sitefinity CMS backend - by clicking  button next to the Show dropdown.
  The generated suggestions can be filtered out per selected AI agent, or all suggestions are displayed together.
• Apply the suggested change directly - by clicking the Apply button, while reviewing the details of a content suggestion.
  The textual change will be persisted/stored in the system when the content editor saves and/or publishes the updated content item.
• Discard the suggested change - by clicking the Dismiss button, while reviewing the details of a content suggestion.
• Preview the suggested change applied inline in the page editor - by clicking Preview in editor.
  Content editors need to confirm if they want to apply or discard the previewed change.
• Restart the tasks performed by the AI agent - by clicking (Refresh AI suggestions).
  This action will possibly result in receiving different content suggestions as the AI agents can adapt and learn from the rejected suggestions.

When you have custom AI agents configured in your Sitefinity CMS backend, multiple AI agents (including the built-in AI agents) can be activated to analyze the same content item. These activated AI agents interact with each other while working in parallel.
In this scenario, AI agents workflow orchestration is introduced to coordinate the execution order of your multiple agents, manage conflicts between them, and prevent repetitive loops of AI-generated suggestions.

The AI agents workflow logic in Sitefinity CMS is based on the Microsoft Agentic framework. Note that Administrators cannot configure a different LLM to be used by the custom AI agents.

In this scenario, some of the generated AI suggestions can overlap so they are aggregated and merged before being surfaced. On the user interface, these AI suggestions visually indicate which AI agents contributed to them.

AI agents execution order

When multiple AI agents are involved (activated) in the same task, they are executed in a coordinated order as per the AI agents workflow logic. Following is the sequence in which multiple AI agents are executed in parallel:

• First in order are the custom AI agents
  
• Next in order is the built-in Search engine optimization Agent
• Last in order is the built-in Brand voice and style Agent

The built-in AI agents have priority over custom agents and their suggestions can override previous suggestions generated by the custom agents when conflicts occur. If no conflicts are detected during the interaction between the AI agents, the generated suggestions are then unified/merged.

Sitefinity AI overview