DurableChat

DurableChat: Multi-Turn Durable Conversations

Overview

This sample demonstrates how to make a multi-turn chat session durable using DurableChatSessionClient. Each call to ChatAsync issues a [WorkflowUpdate] against a long-lived Temporal workflow, so conversation history survives worker restarts without any extra persistence code. Three demos run in sequence: multi-turn context carry-over, tool calls, and history retrieval.

• DurableChatSessionClient.ChatAsync — each message is a workflow update, not a bare HTTP call
• Conversation ID maps 1:1 to a Temporal workflow ID — the same ID routes all turns to the same instance
• GetHistoryAsync retrieves the full message log, including tool call and tool result entries
• DurableAIDataConverter.Instance preserves MEAI's $type discriminator across workflow history round-trips
• UseFunctionInvocation() handles the LLM tool-call loop inside the activity — tool results are sent back to the model before the activity returns

Highlights

• Context carry-over without resending history. The second question ("What is the population of that city?") is answerable because the workflow holds the prior exchange. The caller sends only the new message — the workflow supplies the full history to the LLM automatically.
• DurableAIDataConverter is required. Without it, FunctionCallContent, FunctionResultContent, and other AIContent subtypes lose their $type discriminator when serialized into workflow history, causing deserialization errors on replay.
• Tool calls stay inside the activity. UseFunctionInvocation() completes the full request-call-result loop as one Temporal activity. This is the right model when you want simplicity — see the DurableTools sample if you need per-tool retry policies.
• Sessions have a TTL. opts.SessionTimeToLive controls how long the workflow waits for a new turn before shutting down. Set it to match your expected idle window.

Getting Started

Prerequisites

• .NET 10 SDK or later
• A local Temporal server: temporal server start-dev
• An OpenAI-compatible API key

Configure API credentials

dotnet user-secrets set "OPENAI_API_KEY" "sk-..." --project samples/MEAI/DurableChat
dotnet user-secrets set "OPENAI_API_BASE_URL" "https://api.openai.com/v1" --project samples/MEAI/DurableChat

Run

dotnet run --project samples/MEAI/DurableChat/DurableChat.csproj

Expected Output

Worker started.

════════════════════════════════════════════════════════
 Demo 1: Multi-Turn Conversation
════════════════════════════════════════════════════════
 Conversation ID: multi-turn-<guid>

 User : What is the capital of France?
 Agent: The capital of France is Paris.

 User : What is the population of that city?
 Agent: Paris has a population of approximately 2.1 million ...
════════════════════════════════════════════════════════

(Demo 2 — tool call exchange — output omitted for brevity; shows the model invoking get_weather and returning a forecast.)

════════════════════════════════════════════════════════
 Demo 3: History Query
════════════════════════════════════════════════════════
 Persisted history:
   [User ] Name three planets in our solar system.
   [Agent] Mercury, Venus, and Earth are three planets ...
   ...
 Total messages stored: 4
════════════════════════════════════════════════════════