> For the complete documentation index, see [llms.txt](https://academy.decisionrules.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://academy.decisionrules.io/rule-types/ai-agent-rules.md).
# AI Agent Rules
## Introduction
DecisionRules features two core AI capabilities:
* **AI Assistant:** For authoring and navigating rules faster, supporting your business rule construction.
* **AI Agent Rule:** For the integration of Large Language Model (LLM) technology into precise and deterministic decision-making flows.
In this section we will focus on the second one, you can find more information about the [AI Assistant](https://docs.decisionrules.io/doc/ai-assistant/about-ai-assistant) in our documentation.
While current LLMs work with probability predictions and so can lead to 'hallucinations', Decision Engines are designed for deterministic outcomes and strong auditing. Therefore, for effective integration with our reliable rules, DecisionRules constrains LLMs to an input/output model with a precise structure to guarantee an explainable and predictable business rule.
The **AI Agent Rule** able to handle unstructured tasks— language reasoning, document analysis, handling ambiguity— otherwise not covered by more rigid rules. It transforms unstructured or complex data into **typed, structured JSON responses**, ensuring that even when you use AI, the final outcome remains integrated into a predictable and stable Decision Flow.
{% hint style="info" %}
You can always use the AI Agent by itself, and it is the main advantage, because you define all the logic in natural language without explicit conditional sentences. However, it is a good practice to use an AI Agent to handle the "messy" data at the start of a process, and then pass its structured JSON output to other rules for a final, 100% deterministic result.
{% endhint %}
## AI Agent Rule Designer
The AI Agent Designer is optimized to help you bridge the gap between natural language instructions and clear decisions. The interface is split into two primary areas: the **Left Sidebar** for configurations and the **Main Panel** for building your logic.
### Left Sidebar
The sidebar handles the environment in which your model operates. To set up this background you configure the rule sources and the optional features:
* **AI Model:** Select the specific Large Language Model (LLM) you wish to use (e.g., GPT-4, Claude, etc.).
* **Connector:** Manages the authenticated link between DecisionRules and the AI provider, ensuring secure communication.
* **Cache AI Response (Toggle):** A toggle to enable [caching](https://docs.decisionrules.io/doc/rules/ai-agent/caching). This ensures that identical inputs return the same result instantly, reducing costs and increasing speed for repeated requests.
* **Explainable AI (Toggle):** Enables the generation of a system-defined [explanation text](https://docs.decisionrules.io/doc/rules/ai-agent/explainable-ai), providing transparency into why the model reached a specific conclusion.
* **Data Dictionary:** A live reference panel of all available data. This allows you to drag and drop *Input variables*, *Rule Variables*, and *Attachments* directly into your instructions.
### The Main Panel
This panel is the blank notebook where you describe the plot of your rule. It is organised into four specialized bookmarks:
* **Prompt / Instruction:** The core of the rule. Here, you write natural language instructions telling the AI what to do. You can use {*variables*} from your Data Dictionary to inject real-time data into your prompt.
* **Annotations:** This is the most critical tab for deterministic results. Here, you define exactly what the output JSON should look like, providing data types (Text, Number, Boolean, etc.) and descriptions for every field the model returns.
* **Explainable AI:** Active only when toggled in the sidebar, this tab lets you configure four fields that help auditing the model's logic: `probability`, `reason`, `source_fragments`, and `warnings`. These fields are injected automatically.
* **Attachments:** Allows you to upload reference documents (PDFs, TXT, etc.). These files become part of the rule's versioned logic, allowing the AI to "read" specific policies or guidelines before making a decision.
## Connectors
DecisionRules Spaces allow you to connect directly and more efficiently with your AI provider. These AI models that are set by a secure communication as authentizise provider for all the rules in the Space are counted as [**Connectors**](https://docs.decisionrules.io/doc/space/connectors), it means, you can manage them along side Data and Integration providers.
We support a wide range of relevant Large Language Models for our connectors and continuously expand this list based on client requests.
**The most important aspect of implementation is as follows:** "*To use a model in the AI Agent Rule, you must first connect your LLM to your DecisionRules Space*".
#### Steps to connect your LLM
1. Go to left-side menu and click on **Space**
2. On the new menu opened next to it, select **Connectors**
3. Click on the main purple frame: "Add Connector"
4. Select among the AI models options
5. Fill in the information required
6. Press **Create**
{% hint style="info" %}
You can also configure the connection of a new model within the **Left sidebar** of the *AI Agent Designer*, just click the **+ Create** button next to Connectors.
{% endhint %}
With the tutorial below you will be able to create a simple AI Agent Rule and discover how to integrate the new funcionalities within a Decision Flow.
{% content-ref url="/pages/zUvVqvVlKPzXlvtuk9No" %}
[Create a Simple AI Agent Rule](/rule-types/ai-agent-rules/create-a-simple-ai-agent-rule.md)
{% endcontent-ref %}