Project Overwatch

Project Overwatch is a highly autonomous, multi-agent IT operations swarm. It is designed to intelligently monitor, diagnose, strategize, and resolve infrastructure and application incidents in real-time. Built upon a microservices architecture and advanced Agentic AI paradigms, Overwatch orchestrates automated failovers, service lifecycle management, and Azure infrastructure manipulation through the Model Context Protocol (MCP).

---

🏗️ Architecture Diagram

graph TD
    classDef agent fill:#f9f0ff,stroke:#d8b4e2,stroke-width:2px,color:#000;
    classDef app fill:#e1f5fe,stroke:#4fc3f7,stroke-width:2px,color:#000;
    classDef infra fill:#fff3e0,stroke:#ffb74d,stroke-width:2px,color:#000;
    classDef llm fill:#e8f5e9,stroke:#81c784,stroke-width:2px,color:#000;

    subgraph Target_Environment [🎯 Target Environment]
        App["FastAPI Demo Service"]:::app
        DB[("SQLite DB")]:::app
        App --- DB
    end

    subgraph AI_Agent_Swarm [🤖 Agent Framework & Event Bus]
        S["Sentry Agent<br/>(Monitors Health)"]:::agent
        D["Diagnostician Agent<br/>(Root Cause Analysis)"]:::agent
        ST["Strategist Agent<br/>(Recovery Strategies)"]:::agent
        O["Orchestrator Agent<br/>(Governance & Approval)"]:::agent
        L["Logistics Agent<br/>(MCP Executor)"]:::agent
        TC["Traffic Controller<br/>(Regional Routing)"]:::agent
    end

    subgraph Intelligence [🧠 LLM]
        LLM["FoundryClient / Large Language Model"]:::llm
    end

    subgraph Infrastructure [☁️ Azure & MCP]
        MCP["MCP Tools<br/>(Deploy, Restart, Switch)"]:::infra
        AZ["Azure Infrastructure<br/>(East / West Regions)"]:::infra
    end

    App -- Metrics & Health Logs --> S
    S -- Incident Detected --> D
    D <--> |Analysis Prompt| LLM
    D -- Incident Classified --> ST
    ST <--> |Strategy Prompt| LLM
    ST -- Strategy Ready --> O
    O -- Execute Strategy --> L
    O -- Failover Region --> TC
    L -- Tool Invocation --> MCP
    TC -- Reroute --> MCP
    MCP -- Infrastructure Actions --> AZ

Loading

---

🤖 The Overwatch Agent Swarm

The system's operational intelligence is distributed across specialized roles, communicating asynchronously via an internal event bus:

• Sentry Agent: The monitoring vanguard. It continuously polls the target application's health and telemetry endpoints (/health and /metrics). Upon detecting anomalies—such as elevated latency, increased error rates, or service disruption—it registers an Incident and alerts the overall system.
• Diagnostician Agent: The analytical core. It ingests incident metrics and leverages an LLM (via FoundryClient) to perform root cause analysis, assign structural classifications (e.g., database failure, network outage), and recommend immediate investigatory or remedial actions.
• Strategist Agent: The strategic planner. It synthesizes the Diagnostician's assessment to formulate a concrete, executable recovery strategy (e.g., failover_region, restart_service, scale_service) using AI inference.
• Agent Orchestrator: The governance layer. It evaluates proposed recovery strategies alongside their AI-generated confidence scores. High-confidence strategies (>85%) are auto-approved for execution, while lower-confidence plans trigger requests for human-in-the-loop intervention.
• Logistics Agent: The execution arm. It is responsible for translating approved recovery strategies into tangible, infrastructure-level actions using specific MCP tools.
• Traffic Controller Agent: The network specialist. It is dedicated to managing traffic routing and executing regional failovers when severe, multi-region degradation is detected.

---

🛠️ Model Context Protocol (MCP) Integration

The agent swarm interacts with the underlying infrastructure using specific MCP scripts located in mcp/tools/. These tools provide a standardized interface for complex operational tasks:

• deploy_container.py: Scales services or pushes container updates.
• restart_service.py: Executes lifecycle restarts for Docker or Azure container apps.
• switch_traffic.py: Reroutes traffic between Azure primary (East) and secondary (West) regions via the Azure CLI.
• azure_mcp_server.py: The integration server responsible for handling tool dispatching and execution against Azure resources.

---

🖥️ The Demo Service Environment

A lightweight, fully functional environment designed specifically to be monitored, stressed, and auto-healed by the agent swarm.

• Backend (FastAPI, Python): Located in mini_app/backend/. It features a SQLite database, user CRUD routes (/api/users), and critical monitoring endpoints (/health, /metrics, /status).
• Chaos Engineering Capabilities: Exposes secure administrative endpoints (/admin/degrade, /admin/outage, /admin/recover) to artificially inject latency and errors, facilitating robust testing of the agents' response times.
• Frontend (HTML/JS): Served statically by FastAPI, providing a visual interface for the demo application.

---

☁️ Infrastructure & Deployment

• Local Containerization: Managed via docker-compose.yml, deploying the demo service alongside the agent swarm in isolated, yet connected, containers.
• Azure Infrastructure (Terraform): Located in infra/azure/. It provisions an Azure Resource Group, Container Registry, Log Analytics Workspace, and dual-region (East/West) Container App Environments to support the agents' multi-region failover automation.

---

🚀 Getting Started

1. Configure Environment: Set up your .env variables, including required Azure credentials and OpenAI/Foundry API keys.
2. Install Dependencies:

   pip install -r requirements.txt

3. Initialize Environment:

   docker-compose up --build

4. Launch Overwatch: The primary system entry point is scripts/start_overwatch.py. Execute this script to initiate agent registration, observe the boot sequence, and trigger manual incident simulations.