Your First Agent Flow (Non-Deterministic)¶

AI agent could have a multiple workflows. However in this example, we will create a simple agent with a single Flow which is a conversation bot.

What is a Non-Deterministic Flow?¶

Non-deterministic flows don't follow a predefined path that executes the same way each time. Instead, the flow's execution is determined at runtime, leveraging AI models and available data to make decisions about next steps. Unlike deterministic processes where every step must be explicitly modeled beforehand, non-deterministic flows adapt dynamically. This approach offers several advantages:

• Faster development: Reduced need for extensive process design and testing
• Improved accessibility: Business users can automate processes without deep technical expertise
• Enhanced personalization: Flows can adapt to individual user needs and contexts
• Greater flexibility: The system can handle complex scenarios that would be difficult to model explicitly

In deterministic systems, processes that can't be fully defined in advance typically require human intervention. Non-deterministic flows can increase automation in these scenarios, such as customer support, content generation, or decision-making tasks that traditionally needed human judgment.

However, non-deterministic flows also present challenges, including less predictable behavior, potentially higher execution costs, and more complex auditing requirements. If the business process is well-known and requires repeatable results, then a deterministic flow is a better choice.

Creating a Conversational Flow¶

Update your Program.cs: Program.cs >

using XiansAi.Flow;
using DotNetEnv;

// Load the environment variables from the .env file
Env.Load();

// name your agent
var agent = new Agent("News Agent");

var bot = agent.AddBot<NewsReaderBot>();

await agent.RunAsync();

Add a conversation bot to the agent¶

Agents can be configured to have multiple workflows (business processes). One type of workflow is a conversation bot. This is a bot that can be used to interact with the users.

To add a conversation bot to the agent, you need to add a new class to the agent project.

NewsReaderBot.cs>

using Temporalio.Workflows;
using XiansAi.Flow;

[Workflow("News Agent:News Reader Bot")]
public class NewsReaderBot : FlowBase
{
    public NewsReaderBot(){
        SystemPrompt = "You are a news reader bot.";
    }

    [WorkflowRun]
    public async Task Run()
    {
        await InitConversation();
    }
}

Notes:

• The [WorkflowRun] and [Workflow] attributes are required to mark the method as the entry points for the workflow. You can see more about the Temporal.io workflow engine here.

• InitConversation is a method that initializes the conversation with the user. It is a method that is provided by the XiansAi.Flow library.

• The InitConversation function is designed for flexibility and supports two usage patterns:

  • Providing a system prompt (string): Users can directly pass a system prompt to define the initial context or instructions for the conversation.
  • Providing a knowledge object: Users can specify a Knowledge object that includes a knowledge name, allowing the conversation to be initialized based on a predefined knowledge base or content source.

Testing Your Setup¶

Run the application requesting to test the configuration:

dotnet run

If no errors occur, your agent is now deployed to the Xians.ai platform.

Note: You will have a Warning which we will fix in the next step.

Log in to the Xians.ai Manager portal and you will see your agent under Agent Definitions.

Click Activate and activate a Singleton Flow Run.

Now in the Messaging playground you can start a conversation with your agent.

You can see the conversation history and the agent's response.

Visualizing the Flow's Logic¶

You may have noted the Warning in the previous step. Also the disabled Visualize button in the portal.

This is because the flow class file is not included in deployed package. Let's fix this.

Add the following in your .csproj file:

<Project Sdk="Microsoft.NET.Sdk">
...

  <ItemGroup>
    <!-- Embed the flow source files -->
    <EmbeddedResource Include="NewsReaderBot.cs">
        <LogicalName>%(Filename)%(Extension)</LogicalName>
    </EmbeddedResource>
  </ItemGroup>

</Project>

Now you can run the agent again and you will see the Visualize button enabled in the portal.

dotnet run

You will see the Visualize button enabled in the portal.

Click the Visualize button and you will see the flow's logic.

Next Steps¶

• Adding Capabilities