AI Assistant

The Ask AI button opens an assistant that has direct access to the same solver the Run button uses. It is available on plans that include AI usage. See Plans and Usage Limits.

It is most useful for the part of the work that is fiddly rather than hard: sweeping a parameter across several values, checking what a field expects, and running a variation without disturbing the case you have set up.

What It Can Do

The assistant has three capabilities, and everything it does is built from them.

A solve started from the conversation counts against your solver allowance exactly as one started from the Run button does, and a sweep counts once per value. The parameter table names the count before you confirm it.

What It Will Not Do

These are deliberate limits, not gaps.

It will not choose for you. A sweep tells you what each value does and which ones cleared the limit you set, both of which come out of the solver. How much margin your enclosure needs is an engineering decision, and it stays yours.

It will not draw charts. The assistant receives the summary metrics from a solve, never the underlying time series. It therefore cannot plot a pressure curve, and if it tried, the curve would be invented. Charts come from the workspace.

It will not produce files. There are no download links in the conversation. CSV and PDF exports are made from the toolbar.

It will not assume your pressure limit, and it cannot skip the confirmation. Every solve needs a maximum allowable pressure to judge against, and the value that gets used is the one you approved on the parameter table. The approval travels with your button press, not with anything the assistant says, so asking it to skip the check, or telling it that you have already agreed, changes nothing. That value is a property of your enclosure, and guessing it would produce a confident verdict about a pack it knows nothing about.

Moving a Result Into the Workspace

Before it solves anything, the assistant puts a parameter table on screen with a Confirm & Run button under it, each row noting where that value came from. Nothing is calculated until you press it, and the assistant cannot press it for you, whatever it is asked. Read the table rather than waving it through, particularly the rows marked workspace panel, which it picked up from your screen rather than from what you typed.

When it does run a solve, it returns a compact result card showing peak pressure, relative pressure, whether the valve opened, when it opened, and the limit it was judged against, with the numbers coloured against that limit.

Always check the Limit line on the card. It is the threshold the verdict was judged against, and it is the one figure that turns a pass into a fail without moving a single pressure. A follow-up question carries forward the set you last confirmed rather than whatever your panel shows by then, and the parameter table names the source of every value before you approve it, so the two should agree. Check anyway: the pressure numbers stand on their own, but the pass or fail wording only means what that line says.

The card carries a View Results on GUI button. Pressing it loads the full result into the main workspace: the parameter panel is backfilled with the values that were used, and the charts are drawn. The button then reads Results Shown on GUI, so you can tell which card is currently on screen. From that point the result behaves like any other, so you can export it, adjust it, or re-run it.

This is an explicit action rather than something that happens automatically, so a conversation running in the background never overwrites a case you are in the middle of setting up.

A card stays applicable for about an hour, then stops. Nothing is being held on our side and then thrown away: the card carries its own parameters in your browser, and the button simply stops accepting them after that. If you try an old card and it has expired, ask the assistant to run it again, or reproduce it from the workspace.

Getting Good Answers

Give it the case, not just the question. The assistant cannot see your parameter panel, so a question about “my setup” means less than one that states the numbers.

Useful shapes of request:

• “30 litre pack, one Demo-NCM-40Ah cell venting, limit 50 kPa. Try the S001 valve with one, two, and three valves and tell me where the peak lands.”
• “What is the valid range for the opening pressure field, and what units does it accept?”
• “Which cell entries are in the database, and how much gas does each one release?”
• “My relative peak is 27 kPa against a 20 kPa limit. What are my options?”

Less useful:

• “Is my valve big enough?” without saying which valve, which pack, or which limit.
• Anything asking for a plot, a file, or a link.

Managing the Conversation

The chat window can be dragged, resized, and minimised, and it stays out of the way of the workspace. Starting a new conversation clears the history, and the tool asks you to confirm first. You can export a conversation as a Markdown file if you want to keep the reasoning alongside your results.

What You Should Know About the AI

The assistant is powered by a third-party large language model. What you type, along with the simulation context it needs, is sent to that provider in order to generate a reply. The in-app disclosure notice states the current arrangement, and the Privacy Policy covers it in full.

Its analysis rests on the same simplified model as the rest of the tool, and it can be wrong in ways that read fluently. It does not constitute professional engineering advice and must not be the sole basis for a safety decision.

Treat it the way you would treat a capable colleague who has not seen your pack: useful for exploring options quickly, not a substitute for your own judgement or for Assumptions and Limits.

Worked Example

Sizing Through the AI Assistant runs a complete selection through the conversation, from first question to applying the result.