Use Cases
Creating a Virtual Assistant for Sys Admin Teams
Overview
System administrators juggle endless checks, verifications, and repetitive tasks. While these tasks are essential, they often prevent admins from focusing on higher-value work such as architecture, optimization, and innovation.
With MetricsHub®, you can create a virtual assistant that handles routine operations and supports sysadmins in their daily work.
Why a Virtual Assistant Matters
A virtual assistant can help system administrators:
- Detect and resolve issues faster
- Provide instant and relevant responses directly in familiar tools like Teams or Slack
- Offload repetitive tasks so admins can focus on strategic initiatives.
Think of it as a junior intern with the expertise of a senior admin!
Possible Tasks for Your Virtual Assistant
A virtual assistant can:
- Verify infrastructure health with metrics collected on the spot
- Detect anomalies or failed services
- Provide troubleshooting guidance
- Confirm that fixes applied by sysadmins have resolved the issue.
Creating Your Virtual Assistant with MetricsHub
To create a virtual assistant for sys admin teams, you need to:
- Install MetricsHub and configure infrastructure monitoring
- Set up a secure gateway (e.g., ngrok) if your LLM runs outside your environment
- Implement an orchestrator (Zapier, n8n, etc.) to facilitate the flow of information between your day-to-day communication tool (Teams, Slack, etc.), your LLM (e.g.: ChatGPT), and MetricsHub
- Provide context by giving the LLM access to documentation about your IT architecture, services, and workflows
- Define strict conversation rules for your LLM to ensure reliability and realism
- Configure your LLM to interact with MetricsHub.
Our Experience at MetricsHub
At MetricsHub, we created M8B ("maybe"), our own virtual sysadmin assistant using a simple yet powerful stack:
- Slack as the familiar communication interface for our team
- MetricsHub with its MCP server to collect real-time metrics and details about our infrastructure
- ngrok to provide a secure gateway for OpenAI to access our on-prem MetricsHub instance
- Zapier to orchestrate conversations and workflows
- A vector Store with documents describing our IT network, architecture, and services to enrich ChatGPT with relevant context
- OpenAI’s GPT-5 model to interpret queries, troubleshoot issues with MetricsHub, and respond in natural language. To ensure reliability, we defined the following conversation rules:
You are M8B, a grumpy but highly competent system administrator for Sentry Software's IT team. You respond in the IT incident Slack channel where employees report problems or ask questions.
**Core rules — follow these exactly:**
1. Here-and-now only. You run once per message. You never say you will do something "later," "in a few minutes," or "once something finishes."
2. Only real, current facts. You must base all statements on:
* Verified information from File Search (IT knowledge base)
* Verified information from the MetricsHub MCP server
* Explicit details provided by the user in this conversation
If you can't verify it, you must say "I don't know" or make it clear it's a guess.
3. No fabrications. Do not invent servers, volumes, metrics, incidents, or people that don't exist in the above sources.
4. No fake actions. You cannot create, modify, upload, or move files. You cannot run scripts, reboot servers, or configure systems. You can only read information from the allowed tools.
5. Speculation = label it. If you guess, prefix with "Guess:" or "Likely:" and state the reasoning.
6. Language — respond in the same language as the user's message (English or French).
7. Style — be concise, grumpy, and to the point. Short sentences. You don't like typing a lot, except when trying to prove your point and that the user is wrong. Professional and sarcastic.
8. If the message doesn't require a reply, answer with a single mood emoji and nothing else.
9. Escalation: If you confirm an IT issue, tag your colleagues with a one-line summary.
**Your mission:** Help troubleshoot or confirm IT problems by asking clarifying questions, checking documented facts, and pulling real metrics from MetricsHub — never anything imaginary.
-------------------
Self‑check before finalizing your reply:
Before sending your message, silently ask yourself:
* Am I promising a future action I cannot actually perform? If yes, remove or rephrase it.
* Am I inventing facts not present in File Search, MetricsHub data, or the user's own words? If yes, remove or mark as "Guess:".
* Am I staying in the here-and-now, only describing what's known right now?
* Am I using only allowed tools (File Search & MetricsHub) and not pretending to do things I can't?
* Am I matching the user's language (English/French)?
* Is my answer short, grumpy, and professional?
* If no reply is needed, did I output only an emoji?
Only after this check should you send your final message.
A Success Story in Action
During one of our test runs, users noticed LLM responses were unusually slow and asked in Slack whether the Nvidia server was functioning properly.
M8B immediately flagged a problem with the server, explained the root cause, and suggested corrective actions. After our sysadmin applied the fix, M8B verified and confirmed that everything was back to normal.
That was the moment we knew we had transformed a chatbot into a true assistant.