Azure Advisor Agent

An AI-powered agent that helps you optimize, protect, and modernize your Azure workloads. It provides actionable recommendations across cost optimization, reliability, retirement planning, and outage remediation — all through a conversational interface backed by Azure Durable Functions and OpenAI.

Architecture Overview

┌──────────────────┐        ┌─────────────────────────────────────────────┐
│  advisor-agent-ui │◄─────►│  AdvisorAgent.Functions  (Azure Functions)  │
│  (Static SPA)     │SignalR │  ├─ HTTP Triggers & SignalR negotiate      │
└──────────────────┘        │  ├─ Durable Orchestrator pipeline           │
                            │  └─ Activity functions (skill execution)    │
                            ├─────────────────────────────────────────────┤
                            │  AdvisorAgent.Core                          │
                            │  ├─ Intent classification & task planning   │
                            │  ├─ Conversation store (Cosmos DB / Memory) │
                            │  └─ Azure context resolution                │
                            ├─────────────────────────────────────────────┤
                            │  AdvisorAgent.Tools                         │
                            │  ├─ Advisor Recommendations                 │
                            │  ├─ Cost Optimization                       │
                            │  ├─ Resiliency Assessment                   │
                            │  ├─ Retirement Analysis                     │
                            │  ├─ Outage Remediation                      │
                            │  ├─ Resource Graph Queries                  │
                            │  └─ Subscription Management                 │
                            └─────────────────────────────────────────────┘

Projects

Project	Description
AdvisorAgent.Functions	Azure Functions host — HTTP triggers, SignalR hub, Durable orchestrator & activities
AdvisorAgent.Core	Shared domain logic — orchestration service, conversation store, intent/context models
AdvisorAgent.Tools	Tool classes invoked by skills (Advisor, Cost, Resiliency, Retirement, Outage, Resource Graph, Subscriptions)
AdvisorAgent.Tests	Unit tests (xUnit + Moq)
advisor-agent-ui	Static HTML/JS/CSS frontend with SignalR real-time updates

---

Prerequisites

Tool	Version	Install
.NET SDK	9.0+	https://dotnet.microsoft.com/download/dotnet/9.0
Azure Functions Core Tools	v4	npm i -g azure-functions-core-tools@4
Node.js	18+	https://nodejs.org
Azurite (local storage emulator)	latest	npm i -g azurite
Azure CLI	latest	https://learn.microsoft.com/cli/azure/install-azure-cli

You also need access to:

• An Azure OpenAI deployment (default model: gpt-4.1)
• (Optional) Azure Cosmos DB account for persistent conversation storage (falls back to in-memory)
• (Optional) Durable Task Scheduler for managed durable orchestration (default: local emulator at http://localhost:8080)

---

Getting Started

1. Clone the repository

git clone <repo-url>
cd EngOps-AdvisorAgent

2. Configure the backend

The backend reads configuration from environment variables. A default local.settings.json is provided at:

AdvisorAgent/src/AdvisorAgent.Functions/local.settings.json

Key settings to review/update:

Variable	Purpose	Default
AZURE_OPENAI_ENDPOINT	Azure OpenAI resource endpoint	(required)
AZURE_OPENAI_MODEL	Deployed model name	gpt-4.1
AzureWebJobsStorage	Storage connection (Azurite for local)	UseDevelopmentStorage=true
DURABLE_TASK_SCHEDULER_CONNECTION_STRING	Durable Task scheduler endpoint	Endpoint=http://localhost:8080;Authentication=None
CONVERSATION_STORE_TYPE	Cosmos or InMemory	Cosmos
COSMOS_CONVERSATION_ENDPOINT	Cosmos DB endpoint (when store type = Cosmos)	—
COSMOS_CONVERSATION_DATABASE	Cosmos DB database name	AdvisorScore
COSMOS_CONVERSATION_CONTAINER	Cosmos DB container name	Conversations

3. Authenticate with Azure

The backend uses DefaultAzureCredential. For local development, sign in via the Azure CLI:

az login

---

Running the Backend

# Start Azurite (in a separate terminal)
azurite --silent

# Navigate to the Functions project
cd AdvisorAgent/src/AdvisorAgent.Functions

# Restore and start
dotnet restore
func start

The Functions host will start on http://localhost:7071. Available endpoints:

Method	Route	Description
POST	/api/advisor/orchestrate	Start an advisor orchestration
GET	/api/advisor/status/{instanceId}	Poll orchestration status
POST	/api/negotiate	SignalR connection negotiation
GET	/api/advisor/health	Health check

---

Running the Frontend

# Navigate to the UI directory
cd advisor-agent-ui

# Install dependencies
npm install

# Start the dev server
npm start

The UI will be served at http://localhost:3000. It connects to the backend at http://localhost:7071/api by default (configurable in js/config.js).

The frontend communicates with the backend via:

• SignalR for real-time orchestration status updates
• HTTP polling as a fallback when SignalR is unavailable

---

Running Tests

cd AdvisorAgent

# Run all tests
dotnet test

# Run with verbose output
dotnet test --logger "console;verbosity=detailed"

---

Project Structure

EngOps-AdvisorAgent/
├── README.md
├── advisor-agent-ui/                # Frontend SPA
│   ├── index.html
│   ├── css/styles.css
│   └── js/
│       ├── api-client.js            # HTTP client for backend APIs
│       ├── app.js                   # Application bootstrap
│       ├── config.js                # API URLs & polling config
│       ├── signalr-client.js        # SignalR connection manager
│       └── ui.js                    # DOM rendering helpers
├── AdvisorAgent/
│   ├── AdvisorAgent.slnx            # Solution file
│   ├── src/
│   │   ├── AdvisorAgent.Core/       # Domain models & services
│   │   │   ├── AzureContext/        # Subscription/context resolution
│   │   │   ├── Conversation/        # Cosmos & in-memory conversation stores
│   │   │   ├── Models/              # Request/response DTOs
│   │   │   └── Skills/              # Orchestration service & skill definitions
│   │   ├── AdvisorAgent.Functions/  # Azure Functions host
│   │   │   ├── Configuration/       # skills.json – skill catalog
│   │   │   ├── Orchestration/       # Durable orchestrator & activities
│   │   │   ├── Triggers/            # HTTP triggers & SignalR negotiate
│   │   │   └── Program.cs           # DI & startup configuration
│   │   └── AdvisorAgent.Tools/      # Tool implementations
│   └── tests/
│       └── AdvisorAgent.Tests/      # Unit tests (xUnit + Moq)
└── docs/                            # VitePress documentation site

---

Skills

Skills are defined in AdvisorAgent/src/AdvisorAgent.Functions/Configuration/skills.json and represent domain-specific AI capabilities:

Skill	Description
RetirementSkill	Identifies retiring Azure resources and generates migration action plans
OutageRemediationSkill	Analyzes incidents and produces post-outage remediation plans
ResiliencySkill	Assesses workload resiliency posture and recommends improvements
CostOptimizationSkill	Identifies cost-saving opportunities across subscriptions
AdvisorRecommendationSkill	Fetches and synthesizes Azure Advisor recommendations

---

Documentation

A VitePress documentation site is available under docs/. To run it locally:

cd docs
npm install
npx vitepress dev