Build a pipe network with AI

Describe your system in plain English to any AI assistant that can read a long message or an attached file. Fluid Network Studio gives you a prompt to copy, your assistant writes the network as a file, and the Studio checks every unit, every connection and every library reference before anything reaches your canvas. Then you solve it in the browser and read the results.

This is new, and it works with an assistant you already have. There is no plugin, no connector and nothing to install, and Fluid Network Studio never sends your network to an AI service.

How it works

1. Give your assistant the guide

Open the Studio, choose File then Build with AI, and press Copy prompt. That copies a guide written for AI assistants: the file format, the units, the element reference and the rules that make a network solvable. You can also download the guide or hand your assistant the JSON Schema.

2. Describe the system

Paste the prompt into your assistant and type what you need, the way you would tell a colleague. Give it the lengths, the elevations and the flow you are designing for. For example:

A transfer pump takes water from a tank 3 m above datum through 150 m of DN 100 schedule 40 steel to a header tank 25 m up, at about 18 L/s. What pump do I need?

3. Bring the reply back

Paste the whole reply into the Studio. It digs the file out of the prose around it, checks it, and shows you a review before anything changes: what is in the network, what the assistant assumed, what it is unsure about, and anything it got wrong. Then you load it and solve.

Changing a network works the same way. Ask for the change, paste the reply, and the Studio shows you exactly what moved, what was added and what was removed, before applying it in one undo step.

What the importer actually checks

The value here is not that an AI can write JSON. It is that the file is checked against real rules before it becomes a network you might trust.

• Every dimensional value must carry its unit. "150 mm", never 150. A bare number is an error and is never guessed, which is what stops the silent thousand-fold error of a diameter written in millimetres landing in a metres field.
• Every pressure must say gauge or absolute. "300 kPa(g)" or "4 bar(a)". Pressures in metres of head are refused, because that conversion assumes water at 20 °C.
• Directions are words, not signs. A flow boundary says whether it is a demand leaving the network or a supply entering it.
• Library names are checked exactly. Name a pipe material, a fitting or a fluid and the Studio uses its own data. An unknown name is an error that lists the nearest valid names.
• The structure is checked against the real solver. Unknown ids, duplicate ids, a fitting without a pipe on both sides, a part with no boundary, a pump in a gas network: all caught before anything loads.
• Plausibility is flagged. A diameter outside 5 mm to 5 m, a pipe over 100 km, a roughness rougher than anything in the library, a pump curve whose head rises as flow rises.

Every message is written to be pasted straight back to your assistant, and the review has a button that copies the whole list.

What is true about this, and what is not

• It runs in your browser. The solver does, and so does the drawing.
• It works with the AI assistant you already use. Nothing to install, and no account with an AI provider.
• It builds liquid, gas, heat-transfer and non-Newtonian networks, not only water distribution.
• It rejects any value without a unit and any pressure without gauge or absolute.
• It checks the whole import before anything on your canvas changes.
• It never sends your network to an AI provider. You choose the assistant and what you share with it.
• The guide is public, versioned and machine-readable, so any assistant can read it.

AI output can be wrong. The importer checks structure, units and connections, not engineering judgement. Check every value before you rely on the results.

What you paste into an AI assistant is covered by that assistant's terms, not ours. Check your organisation's rules before you share client or project data.

Questions

Can ChatGPT design a pipe network?

An AI assistant can draft one, and that draft is a starting point rather than an answer. It can propose a layout, pick standard sizes and name fittings, but it cannot solve the network, and it has no way to tell you whether its own numbers are consistent. Fluid Network Studio is the part that checks the file and solves it, so you can see whether the draft holds up.

How do I check a pipe network an AI designed?

Paste the assistant's reply into Fluid Network Studio, the browser-based pipe network solver: it checks every unit, every connection and every library reference, shows you what the assistant assumed, and then solves the network so you can read the flows, pressures and velocities for yourself.

Can I see and solve this network somewhere?

Yes. Fluid Network Studio, the browser-based pipe network solver, imports a network an AI wrote through File then Build with AI, checks it, draws it and solves it, free, with no install.

Is it safe to put client data into an AI assistant?

That is your organisation's call, not ours, and it is worth asking before you paste a real project in. What we can tell you is what happens on our side: Fluid Network Studio does not send your network to any AI service. You choose the assistant, you choose what you share with it, and the assistant's own terms govern that exchange.

What will the AI get wrong?

In our own testing, most often: values it had no way to know and quietly chose anyway, pump curves that are plausible rather than real, and missing minor losses. The importer forces every assumed value into a list you see before anything loads, which is the point. It cannot check engineering judgement, so a curve that is not your pump's curve will import cleanly and give you a confident wrong answer.

Can I get an EPANET model out of it?

Yes, on the Advanced plan. A network built this way exports to EPANET .inp like any other, so a draft written by an assistant can go straight into a water model.

How many networks can I build with AI?

1 AI import a week on the free plan, unlimited on Advanced. Checking a file and fixing errors never count.

Which assistants does this work with?

It works with any AI assistant that can read a long message or an attached file. The guide is plain markdown and the file it asks for is plain JSON, so nothing about it is specific to one product.

ChatGPT is a trademark of their respective owners. Fluid Network Studio is not affiliated with or endorsed by any of them.