persisted-conversation

Persisting and Resuming Agent Conversations

::: zone pivot="programming-language-csharp"

This tutorial shows how to persist an agent conversation (AgentThread) to storage and reload it later.

When hosting an agent in a service or even in a client application, you often want to maintain conversation state across multiple requests or sessions. By persisting the AgentThread, you can save the conversation context and reload it later.

Prerequisites

For prerequisites and installing NuGet packages, see the Create and run a simple agent step in this tutorial.

Persisting and resuming the conversation

Create an agent and obtain a new thread that will hold the conversation state.

using System;
using Azure.AI.OpenAI;
using Azure.Identity;
using Microsoft.Agents.AI;
using OpenAI;

AIAgent agent = new AzureOpenAIClient(
    new Uri("https://<myresource>.openai.azure.com"),
    new AzureCliCredential())
     .GetChatClient("gpt-4o-mini")
     .AsAIAgent(instructions: "You are a helpful assistant.", name: "Assistant");

AgentThread thread = await agent.GetNewThreadAsync();

Run the agent, passing in the thread, so that the AgentThread includes this exchange.

// Run the agent and append the exchange to the thread
Console.WriteLine(await agent.RunAsync("Tell me a short pirate joke.", thread));

Call the Serialize method on the thread to serialize it to a JsonElement. It can then be converted to a string for storage and saved to a database, blob storage, or file.

using System.IO;
using System.Text.Json;

// Serialize the thread state
string serializedJson = thread.Serialize(JsonSerializerOptions.Web).GetRawText();

// Example: save to a local file (replace with DB or blob storage in production)
string filePath = Path.Combine(Path.GetTempPath(), "agent_thread.json");
await File.WriteAllTextAsync(filePath, serializedJson);

Load the persisted JSON from storage and recreate the AgentThread instance from it. The thread must be deserialized using an agent instance. This should be the same agent type that was used to create the original thread. This is because agents might have their own thread types and might construct threads with additional functionality that is specific to that agent type.