AI Generator

The AI Generator turns a plain-English description of a network into a proposed set of devices and links. You review the proposal, and ONEP creates the environment on the canvas, stopped, for you to edit and start. It builds topologies; it does not chat, and it never starts, deploys or downloads anything on its own. Open it from AI Generator in the sidebar.

How it works#

A row of five steps across the top shows where you are: Describe, Generate, Review, Deploy and Done.

1. DescribeType what you want in English into the field labeled Describe your network environment..., up to 1,000 characters (a counter shows how many you have used), or select an example card to fill the field. Then choose Generate.
2. GenerateONEP sends your text to the AI source set in Settings › AI and gets back a proposed topology. A counter shows the seconds elapsed. Nothing is created yet, and the browser warns you if you try to leave the page while it works.
3. ReviewCheck the devices and links. ONEP picks an image from your Centrum for each device; you can change it. See Review the proposal.
4. DeployChoose Deploy environment. ONEP creates the environment on the canvas with every device stopped, exactly like one you drew yourself.
5. DoneThe page shows how many devices and links were added. Open editor opens the environment on the canvas; Generate another starts again.

Write a description that works#

• Name concrete devices and counts: "three routers in a triangle running OSPF, with a firewall to the internet".
• Keep it self-contained. The generator creates a new environment; it cannot see or change one you already have.
• If you use an external AI service, the text you type leaves the appliance. Leave out real addresses and names from your own network.

The Examples cards fill the field with a ready description. On Community and Lite the examples are Two-router OSPF, Web + DB on one switch and Firewall between two segments. Pro adds larger ones: GOAD, GOAD-Light, MINILAB, SCCM Lab, NHA, DRACARYS, SOC Training, BGP Peering and OT / ICS. A model can answer differently each time. If the result is not what you wanted, change the description or choose Start over.

Review the proposal#

• The top of the proposal shows its name, description and the number of devices and links.
• Each device has an image list, filled in with the best match from your Centrum. A device with no image is highlighted amber, and Register next to it opens Centrum so you can add one. The list reads — select an image — until you choose.
• Each device also has editable vCPU, memory (MB) and disk (GB) fields. ONEP raises a value that is below the minimum for that kind of device when it creates the environment.
• Total resources adds up vCPU, memory and disk. If the totals exceed what the appliance has free, a warning says ONEP recommends more headroom but does not block you. If you raise the sizes after that check, a second note says the check has not been re-run for the new total.
• Connections lists every link as device:interface ↔ device:interface.
• Deploy environment stays off until every device has an image; the line beside it counts the devices still without one. ← Back returns to the description and Start over clears it.

Limits#

What it refuses#

• Anything that is not a network or security topology. This is a deliberate scope check. The answer reads I only build network topologies. Describe a network — e.g. "a firewall between two PCs" or "3 routers in a ring".
• Other languages. A description in a non-Latin script is refused with The assistant understands English only. A description in Latin letters that is not English is refused with ONEP Assistant works in English. Please describe the environment in English.
• Questions about ONEP itself. The generator will not discuss the platform's internals, configuration or code.
• Too many devices. The answer says how many devices one environment can hold.

It also will not create images, write device configuration, or accept a topology that skips ONEP's checks: every proposal goes through the same device limit, role and link checks.

Set up the AI source#

ONEP does not supply a model account or a key. An administrator chooses where the AI Generator gets its model in Settings › AI; the tab is visible to administrators only. A line at the top of the card shows ONEP Assistant is connected, ONEP Assistant is not connected, Checking the ONEP Assistant... or, when the check does not answer in time, The ONEP Assistant could not be checked. Try again in a moment. The AI source list has two choices:

Save stores the choice.

Model on this appliance#

The model runs on your own hardware. On an appliance without a graphics processor expect roughly ten seconds to two minutes per generation; a GPU is used automatically when one is present. For better accuracy, Download recommended model (2 GB) downloads a larger model and restarts the assistant on it. It needs 4 GB of free memory and takes a few minutes; the button shows its progress, and a message tells you when ONEP Assistant is ready.

External AI service#

1. Base URLThe OpenAI-compatible address of your service, for example https://api.openai.com/v1 or your own gateway.
2. API keyPaste your key. Once saved it is never shown again; the field reads saved, paste to replace.
3. ModelPick from the list once the service answers, or type the model name if no list appears.

Save stays off until a base URL and a model are set. When you choose this source the card warns that prompts leave the appliance and go to your service, on your own account.

Your own model choice#

Each person can pick the model their own requests use under Settings › Account › AI preferences. The Model list starts at Server default and offers the models the administrator allows. If the administrator has limited it to one model the list is read-only, and Server default is always available to switch back.

If it does not work#