UplivraUplivra
Uplivra is in beta and online purchasing is paused. Apply to test it: testers get a license on request and 10% off when purchasing opens.

Home › Install guides › Connect an AI

Install guide · Administrators

Connect an AI

Optional and free. Three ways to use an AI with Uplivra - let an AI app your company already uses read Uplivra, point Uplivra at a model server on your own network, or use a cloud AI service with your own API key - and what each one sends.

Uplivra Technologies LLC · Guide for Uplivra 26.10 · Updated 4 October 2026 · Latest version: https://uplivra.com/guides/connect-ai.html

Download this guide as a PDF

Before you start

Uplivra doesn't need an AI. Insights (questions in plain words, likely causes, notices, Running out forecasts) is Uplivra's own analysis and works without one. Uplivra ships no AI model and there is nothing extra to download.

Connecting an AI is optional and free from Uplivra. Everything below is under Settings › Connect an AI, and only an administrator can change it. There are three ways:

WayWhat it doesDoes data leave your network?
1. Connect your own AIAn AI app your company already uses (Claude, Microsoft Copilot or another MCP app) reads UplivraOnly if that AI app runs in the cloud
2. Your own model serverA model on your network words the answers on Insights › AskNo
3. A cloud AI serviceOpenAI, Anthropic (Claude), Azure OpenAI or Google Gemini words the answers, with your own API keyYes: see What is sent

You can use way 1 together with way 2 or way 3.

Way 1: connect your own AI

Your AI app connects to Uplivra over MCP (Model Context Protocol, a standard way for AI apps to read other software). The connection is read-only: Uplivra gives the AI no way to change devices, settings or alerts. It is off until an administrator turns it on.

Turn it on

  1. Open Settings › Connect an AI.
  2. Tick Allow other AIs to read Uplivra and click Save.

Make a connection (one per AI app, or one per person)

  1. Under Add an AI, pick Which AI: Claude Code or Claude Desktop, Claude (claude.ai Team or Enterprise), Microsoft Copilot Studio, Google Gemini (Antigravity), Open WebUI, or Other MCP app.
  2. Tick What it may read: sites and collectors, devices and their checks, alerts, changes, log messages (syslog), SNMP traps, and questions to Insights. Sites, devices, alerts and changes are ticked to start with. Log messages can contain user names and addresses, so tick them only if you need them.
  3. Give it a Name (for example Copilot for the help desk) and choose how long it works for: 7 days, 30 days, 90 days or 1 year.
  4. Click Create the connection.
  5. Copy the token now. Uplivra keeps only a fingerprint of it, so it can't show it again. Keep it like a password.
  6. Follow the steps Uplivra shows for the AI you picked. Most apps need these three things:
SettingValue
Addresshttps://<your Uplivra server>/mcp
TransportStreamable HTTP
Sign-inRequest header Authorization: Bearer <token>

For Claude Code, Uplivra shows a ready command to paste:

claude mcp add --transport http uplivra https://<your Uplivra server>/mcp --header "Authorization: Bearer <token>"
  1. Ask your AI something like "What's down in Uplivra right now?"

Good to know

  • Each token belongs to the person who made it. The AI never sees more than that person can see in Uplivra, and only what was ticked.
  • AI apps that run in the cloud (claude.ai, Copilot Studio) must reach your Uplivra server from the internet over HTTPS, with a trusted certificate. Apps on your own computers (Claude Code, Claude Desktop, Open WebUI) only need your network.
  • ChatGPT is in the list but can't connect yet: it signs in to MCP servers in a different way (OAuth) that Uplivra doesn't offer yet.
  • Each connection can ask up to 600 questions an hour.
  • Every call is in the audit log (Settings › Audit log, action mcp.call).
  • To stop one AI, click Disconnect next to it under Connected AIs. It stops right away.
  • Customers' own administrators make connections; an MSP can't make them for a customer.

Way 2: your own model server

A model server is a program on one of your computers that runs an AI model. Uplivra uses it to word the answers on Insights › Ask in full sentences, and to read unusually worded questions. Uplivra still finds the facts itself; the model only works with what Uplivra found. Nothing leaves your network.

Any server with an OpenAI-style /v1/chat/completions address works, for example Ollama, LM Studio, llama.cpp's server, vLLM or LocalAI.

  1. Install the model server on a computer on your network and download (pull) a model in it. With Ollama, ollama list shows the models you have. If the model server is on another computer, let it listen on the network (Ollama: OLLAMA_HOST=0.0.0.0).
  2. Open Settings › Connect an AI. Under AI model for Insights, tick Use an AI model.
  3. Under Where the model is, choose Your own model server.
  4. Address: the server's address on your network, for example http://192.168.1.20:11434 (Ollama uses port 11434, LM Studio 1234). It must be a private address; Uplivra refuses addresses on the internet here.
  5. Model: exactly as the server shows it, for example a model you've pulled in Ollama.
  6. API key: only if your model server asks for one. Most don't.
  7. Choose what the model may do: help read the question, write the answer, and whether it may see log message text (passwords and keys are masked either way).
  8. Click Save and test the model. Uplivra sends a one-word test and tells you how long the answer took.

Way 3: a cloud AI service

Uplivra can send questions to OpenAI, Anthropic (Claude), Azure OpenAI or Google Gemini, using an API key from your own account with that company. You pay the provider directly; Uplivra charges nothing for it. It is off by default.

  1. Get an API key from the provider's website. For Azure OpenAI, also note your resource address (like https://acme.openai.azure.com) and the deployment name from the Azure portal.
  2. Open Settings › Connect an AI. Under AI model for Insights, tick Use an AI model.
  3. Under Where the model is, choose OpenAI, Anthropic (Claude), Azure OpenAI or Google Gemini.
  4. Address: leave it empty, except for Azure OpenAI: your resource address.
  5. Model: the model name from your provider (for Azure OpenAI, the deployment name).
  6. API key: paste it. It is stored encrypted on your Uplivra server and never shown again. Switching to another provider removes the saved key.
  7. Read the warning and tick I understand what is sent and want to use this cloud service.
  8. Leave Send device names, site names and addresses as they are unticked (recommended): Uplivra then replaces them with stand-ins such as host-3 before sending, and puts the real names back in the answer you see.
  9. Click Save and test the model.

What is sent

  • Way 1: what the AI reads (only what you ticked) goes to that AI app. If the app runs in the cloud, it goes to its provider.
  • Way 2: nothing leaves your network.
  • Way 3: for each question, the question and the facts Uplivra found go to the provider over the internet: measurements, alerts, and log message text if you allow it. Device names, site names and IP addresses are replaced by stand-ins before sending, unless an administrator chooses to send them as they are. Passwords and keys in log text are masked either way.

Every call is in the audit log (Settings › Audit log): mcp.call for an AI app reading Uplivra (way 1), ai.call for each request to your model server or cloud service (ways 2 and 3).

Turning it off

  • Way 1: in Settings › Connect an AI, untick Allow other AIs to read Uplivra and click Save. Every connection stops at once.
  • Ways 2 and 3: in Settings › Connect an AI, untick Use an AI model and click Save. To also remove a saved API key, tick Remove the saved key.

Insights keeps working without any AI: answers are then written by Uplivra's own analysis.