HoldThatThread

A .NET API for managing reasoning conversations with digression support, powered by Azure OpenAI's advanced models.

Status: On hold
Development is paused.
The repository is kept for reference; issues and PRs are not being actively monitored.

Overview

HoldThatThread enables deep reasoning conversations with AI while supporting quick clarification "digressions" without disrupting the main conversation flow. It uses dual Azure OpenAI deployments: extended thinking models (o3-mini) for main conversations and fast models (gpt-4o-mini) for digressions.

Features

• Streaming Reasoning Responses: Real-time streaming with extended thinking using Server-Sent Events (SSE)
• Digression Mini-Chats: Quick clarification conversations that can be merged back or discarded
• Session Management: Persistent conversation history with turn-based interactions
• EventSource Compatible: Two-step GET SSE pattern for browser EventSource API
• Dual Deployment Strategy: Separate Azure OpenAI deployments optimized for reasoning vs. quick responses
• Flexible Provider Support: Works with both Azure OpenAI and OpenAI API
• Enterprise Ready: Azure Key Vault integration, Managed Identity authentication, and API Management support

Architecture

Built using Clean Architecture principles:

• Domain Layer: Core business entities (Session, Message, ConversationTurn, DigressionSession)
• Application Layer: Business logic and services (ReasoningService, DigressionService)
• Infrastructure Layer: External integrations (Azure OpenAI client, in-memory stores)
• API Layer: RESTful endpoints with OpenAPI documentation

Tech Stack

• .NET 9.0
• Azure OpenAI Service
• Server-Sent Events (SSE)
• Azure Key Vault
• Azure API Management (optional)
• Bicep (Infrastructure as Code)
• xUnit (Testing)

Prerequisites

• .NET 9.0 SDK
• Azure subscription (for production deployment)
• Azure OpenAI Service with deployments (or OpenAI API key for development)

Quick Start

Local Development

1. Clone the repository

   git clone <repository-url>
   cd HoldThatThread

2. Configure settings

   For Azure OpenAI:

   {
     "OpenAIProvider": "AzureOpenAI",
     "AzureOpenAI": {
       "Endpoint": "https://your-resource.openai.azure.com/",
       "ReasoningDeployment": "o3-mini",
       "DigressionDeployment": "gpt-4o-mini",
       "ApiKey": "your-api-key"
     }
   }

   For OpenAI:

   {
     "OpenAIProvider": "OpenAI",
     "OpenAI": {
       "ApiKey": "your-openai-api-key",
       "ReasoningModel": "o3-mini",
       "DigressionModel": "gpt-4o-mini"
     }
   }

3. Run the API

   cd HoldThatThread.Api
   dotnet run

4. Access the API

   • API: http://localhost:5146
   • OpenAPI docs: http://localhost:5146/openapi/v1.json

API Usage

See docs/endpoints.md for detailed API documentation.

Quick Example

// Step 1: Start a conversation turn
const response = await fetch('/api/chat/main/turn', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    sessionId: null,
    userInput: "Explain quantum entanglement"
  })
});
const { turnId, sessionId } = await response.json();

// Step 2: Stream the response
const eventSource = new EventSource(`/api/chat/main/stream/${turnId}`);

eventSource.addEventListener('thought', (e) => {
  const data = JSON.parse(e.data);
  console.log('Thinking:', data.text);
});

eventSource.addEventListener('answer', (e) => {
  const data = JSON.parse(e.data);
  console.log('Answer:', data.text);
});

eventSource.addEventListener('done', () => {
  eventSource.close();
});

Deployment

Azure Deployment

The project includes complete Infrastructure as Code (Bicep templates) for automated Azure deployment.

1. Review infrastructure setup

   cd infra
   cat README.md

2. Deploy to Azure

   cd infra/scripts
   ./deploy.sh

The deployment includes:

• Azure App Service (Linux)
• Azure OpenAI Service with two deployments
• Azure Key Vault for secrets management
• Application Insights for monitoring
• Azure API Management (optional)

See KEYVAULT-SETUP.md and APIM-SETUP.md for additional configuration details.

Testing

The project includes 82 comprehensive tests covering:

• Domain logic
• Application services
• Infrastructure components
• API endpoints

dotnet test

Test Coverage

• Domain: Session management, message handling, digression logic
• Application: Reasoning and digression services
• Infrastructure: Azure OpenAI client integration, storage implementations
• API: Endpoint behavior and request/response handling

Project Structure

HoldThatThread/
├── HoldThatThread.Domain/          # Core business entities
├── HoldThatThread.Application/     # Business logic and services
├── HoldThatThread.Infrastructure/  # External integrations
├── HoldThatThread.Api/             # REST API endpoints
├── HoldThatThread.Web/             # Web UI (optional)
├── HoldThatThread.Tests/           # Test suite
├── infra/                          # Bicep templates for Azure
├── docs/                           # API documentation
└── apim-policies/                  # API Management policies

Configuration

Environment Variables

• OpenAIProvider: "AzureOpenAI" or "OpenAI"
• KeyVault__VaultUri: Azure Key Vault URI (production)
• ASPNETCORE_ENVIRONMENT: Development/Production

Key Vault Secrets (Production)

• AzureOpenAI--Endpoint
• AzureOpenAI--ApiKey
• AzureOpenAI--ReasoningDeployment
• AzureOpenAI--DigressionDeployment

Contributing

1. Fork the repository
2. Create a feature branch
3. Write tests for new functionality
4. Ensure all tests pass
5. Submit a pull request

Contact

Ammons DataLabs Email: jaybea@gmail.com

License

This project is licensed under the MIT License - see the LICENSE file for details.

Copyright (c) 2025 Ammons DataLabs