Introduction

Modern software systems are becoming increasingly complex. A single production incident can involve multiple services, application logs, database failures, deployment issues, and communication between engineering teams.

For DevOps engineers and Site Reliability Engineers (SREs), investigating these incidents often means switching between multiple tools, monitoring dashboards, and communication channels.

We wanted to explore a better approach.

That's how we built DevOps Command Center AI — an AI-powered DevOps incident investigation and orchestration platform integrated with Switch Gateway.

Our goal was to bring AI agents, incident communication, real-time monitoring, and human-controlled operations into one centralized dashboard.

The Problem We Wanted to Solve

Traditional incident management workflows can become fragmented.

Engineers often need to:

Investigate application failures manually.
Analyze logs from different sources.
Coordinate with multiple team members.
Track incident updates across communication channels.
Evaluate remediation recommendations before taking action.

We wanted to create a centralized environment where an AI agent could assist with investigation while human operators retained control over important decisions.

Why We Chose Switch

Switch provides an infrastructure for connecting and coordinating AI agents through a centralized Gateway.

Instead of building every agent communication mechanism from scratch, we used Switch as the communication and integration layer for our application.

Our implementation uses Switch Gateway for:

Agent registration and discovery.
Authenticated communication.
Room-based agent interaction.
Real-time event streaming.
Agent activity monitoring.

This allowed us to focus on building the DevOps experience around connected agents.

Connecting Our AI Agent

For our initial implementation, we connected a DevOps investigation agent named:

devops-investigator

The agent uses the Antigravity CLI connector and is registered with Switch Gateway.

Its intended responsibilities include:

Investigating application incidents.
Analyzing available logs.
Identifying potential root causes.
Recommending corrective actions.

The agent is displayed in our AI Agents dashboard, where operators can inspect its registration details, connector information, and granted tools.

Building the Incident Response War Room

One of the most important parts of our implementation is the Incident Response War Room.

We configured a dedicated room through Switch Gateway where the operator can communicate with the registered AI agent.

Instead of managing agent interactions in a separate interface, we integrated the room directly into our DevOps Command Center.

The workflow allows operators to:

Open the Incident Response Channel.
Send an investigation request.
Dispatch the message through Switch.
Receive agent responses.
Monitor the conversation in the dashboard.

This creates a centralized environment for incident investigation and coordination.

Real-Time Monitoring with Server-Sent Events

Incident investigation is not just about sending messages. Operators also need visibility into what is happening.

We integrated Server-Sent Events (SSE) to receive live updates from Switch Gateway.

Our application uses real-time event streams to monitor agent and chat activity.

The frontend processes these events to update relevant dashboard information without requiring continuous manual refreshes.

This powers the Live Logs experience and helps operators follow activity as it happens.

Human-in-the-Loop Governance

Automation is useful, but allowing an AI agent to execute sensitive production operations without human authorization introduces risk.

We designed our platform around a human-in-the-loop approach.

The application includes an Action Approvals interface intended to provide visibility into approval requests and operator authorization.

Sensitive or destructive actions should require explicit human approval rather than being executed automatically.

This keeps human operators involved in critical operational decisions.

Our DevOps Command Center

We built a centralized web dashboard using:

React
TypeScript
Vite
Tailwind CSS
Python backend integration
Switch Gateway
PostgreSQL
REST APIs
Server-Sent Events

The dashboard includes:

Overview Dashboard: A centralized view of incidents, agent activity, and service health.

AI Agents: A place to discover and inspect registered Switch agents.

Incident Response: A dedicated War Room for operator-agent communication.

Live Logs: Real-time activity monitoring.

Approvals: Human-in-the-loop governance interface.

Workflows: A visual representation of incident investigation and remediation stages.

Deployments: A dashboard for viewing release history and demonstration scenarios.

Some dashboard metrics, cluster health values, and deployment scenarios use simulated demonstration data rather than live production infrastructure.

Challenges and What We Learned

Building this project gave us several practical lessons.

Agent Connectivity

Connecting an agent is only the first step. Managing its session, registration, communication, and availability requires careful integration.

Real-Time Event Handling

SSE integration requires attention to reconnection, event handling, and duplicate event prevention.

Authentication and Security

We used Switch Gateway authentication and session cookies instead of exposing backend credentials in the frontend.

Separating Real and Simulated Data

A monitoring dashboard should clearly distinguish actual integration status from simulated infrastructure metrics. This is important for transparency and trustworthy demonstrations.

What's Next?

There is still room to expand the platform.

Our future plans include:

Connecting additional specialized DevOps agents.
Integrating external collaboration platforms.
Improving incident correlation.
Expanding real infrastructure monitoring.
Testing complete approval and remediation workflows.
Connecting actual deployment pipelines.

We see this project as a foundation for a more collaborative and intelligent DevOps incident response environment.

Conclusion

Building DevOps Command Center AI helped us explore how Switch can serve as a communication and orchestration layer for AI-powered operational workflows.

By combining a registered AI agent, a centralized Incident Response War Room, real-time event streaming, and human-in-the-loop governance, we created a unified environment for DevOps incident investigation.

The biggest takeaway was simple:

AI agents become more useful when they can collaborate through a structured system while humans remain in control of critical decisions.