AI agentInfrastructure
AI systems administrator agent for a managed service provider
An AI agent that watches monitoring alerts, investigates, fixes routine faults within an agreed safe list, and escalates the rest. Built for one UK business, now rolling out across an MSP’s client base.
The problem
A managed service provider’s service desk sees a constant stream of raw monitoring alerts. Many point to routine faults with known fixes, but each one still needs a person to read it, investigate it, and act.
What Evoloop built
- An AI systems administrator agent that watches Zabbix monitoring alerts as they arrive
- A runbook knowledge store in a vector database that the agent uses to investigate each alert
- An agreed safe list of routine faults the agent is allowed to fix on its own
- Escalation into service desk tickets, either for the desk to action or to approve the agent’s proposed fix
- A log of every action the agent takes
How it runs
- Built on the OpenClaw agent framework
- The agent monitors alerts, investigates them against its runbook knowledge store, and resolves routine faults inside the agreed safe list
- Anything consequential becomes a ticket for the service desk to action or to approve the agent’s proposed actions
- Every action is logged
- It began as a deployment for a single UK business; the Evoloop blog post on the AI sysadmin agent tells that original story
Stack
OpenClawZabbixVector database
Where it is now
Adopted by a managed service provider. Rollout across the MSP’s client base is in progress.