qinshihu/itops-agent-platform

What it solves

ITOps Agent Platform addresses the inefficiency and high stress of traditional IT operations. Instead of manual troubleshooting (VPN login, SSH, command-line investigation) that can take 30-60 minutes per incident, this platform automates the entire lifecycle from alert reception to verification, reducing response times to under 3 minutes.

How it works

The platform implements a closed-loop "Sense $\rightarrow$ Diagnose $\rightarrow$ Decide $\rightarrow$ Execute $\rightarrow$ Verify" workflow using a multi-agent AI system:

1. Alert Integration: Receives alerts from tools like Prometheus or Zabbix via Webhooks.
2. AI Diagnosis: Multiple AI agents collaborate to analyze logs and metrics to find the root cause and generate a natural language report.
3. Decision & Approval: The AI generates structured repair commands and risk assessments, which are pushed to mobile apps (WeCom, DingTalk) for human-in-the-loop (HITL) approval.
4. Execution: Once approved, the platform executes the commands via SSH or API and automatically verifies the result.
5. Infrastructure Management: It provides a unified interface for managing Docker containers, VMware/KVM virtual machines, Kubernetes clusters, and physical data center assets (including 3D digital twins).

Who it’s for

• Operations Engineers & SREs: To eliminate midnight manual troubleshooting and tool-switching.
• IT Managers/CTOs: To move from reactive "luck-based" recovery to autonomous self-healing strategies.
• Small to Medium Enterprises: As a free, open-source alternative to expensive commercial suites like PagerDuty, Rundeck, and Portainer.
• Security Teams: To ensure all repair actions are audited and approved.

Highlights

• Multi-Agent Orchestration: Uses 12 specialized AI agents for collaborative reasoning rather than a single LLM call.
• Full-Loop Automation: Engineering implementation of the entire path from alert to verification.
• Human-in-the-Loop (HITL): Integrated mobile approval flow to ensure safety in production environments.
• Unified Infrastructure Control: Combines container, VM, K8s, and network device management in one platform.
• Command Safety Engine: Includes safety policies and role-based access control to filter dangerous AI-generated commands.
• Local AI Support: Supports local LLM deployment to keep sensitive operational data within the corporate perimeter.