I was just looking to plug a new zigbee coordinator into a POE switch today, then thought - that’s going to be on the wrong VLAN, or, did I trunk the uplink…?
So… that made me think - is anyone using / tried / failed to export device configs, feed it into an AI, and get it to draw out the network??
Nope, and I wouldn’t recommend it either. They are generally not great with mapping IMO, no matter the model.
And really this is better managed with a quick diagram and a spreadsheet. If you want some software, look at netbox for this, you can add in validation rules and all kinds of automations as well.
But really, use a spreadsheet.
I have Visio drawings already, and KeePass files and backup configs and…
So, it’s mostly there, but there’s always something that’s not captured, like maybe MAC address I’m trying to resolve in a switch…
I was thinking AI could draw, redraw and draw again…
You can, I’m just saying be prepared for disappointment. The amount of effort you need to put in, imo, you could have had a spreadsheet and been done. And then used that with a script to autogenerate some mermaidml to get a purdy drawing out of it (which I highly recommend leveraging Beautiful-Mermaid ).
And then you can feed that to an llm and make your queries about ports or other uses, to me thats a much better use/approach. To each their own though.
Not really seen much of mermaid diagrams before, thanks.
Btw, Looking at the issue comments, seems beautiful mermaid is unmaintained and now forked to zombie mermaid: https://github.com/dfadler/zombie-mermaid
Yes and no. I use llama server to run my own local models, and then open ice to monitor it. From there a lot of open source services host MCP servers which then can help integrate. I then use it to auto write the documentation on the layout so next time when I say “go to the ZigBee controller and tell me…” It knows first go to the docs, it remembers where it is and any useful commands, then continues on.
Ah, ok, so you’re usng it as a live agent to interact with the live system as / when you need some info?
Hmm, not thought of doing live queries…
I don’t have any AI running, hence just thinking of using public LLMs to create a drawing, but maybe this can interact with uptime kuma (for example) and tell me more about what just broke…


