AI disclosure

While25 Desktop includes an AI assistant. You describe what you want in plain words and it writes Python: an indicator, a panel, a strategy, a report. It can read the files in your workspace, the chart you are looking at and the panels you have open, so that what it writes fits the work in front of you.

It can also act inside the App: change charts, run the Python it wrote on your computer, fill in an order ticket for you to submit, and cancel an order or stop a strategy after you confirm in a dialog.

It is a tool for building your own tools. It is not a trader, not an adviser, and not a source of signals.

This is a property of AI systems generally, not a defect specific to ours. In particular:

• It can produce code that runs but does the wrong thing.
• It can misread your intent, or answer a question next to the one you asked.
• It can state something confidently that is not true, including about market mechanics, an instrument, or an API it is calling.
• It can use a function, parameter or library that does not exist.
• It can be inconsistent: the same question twice can give two different answers.
• It has no knowledge of anything after its training data except what is in your workspace and the context it is given.

We do not warrant that the assistant is accurate, complete, or fit for any particular purpose. See the licence and the risk disclosure.

These are design limits in the product, not promises about good behaviour:

• It does not decide what you trade. It does not recommend securities, and nothing it writes is investment advice.
• It cannot place an order. It can fill in the order ticket; only you submit it.
• It cannot start a live strategy. Going live takes an explicit confirmation from you, every time.
• It asks before it cancels or stops. Cancelling an order or stopping a strategy waits for your confirmation each time.
• It does not change a running strategy on its own. Once your rules are running, orders come from those rules.
• It shows its work. The steps it took are listed, and the code it wrote is yours to read, edit or delete before anything runs.

The assistant currently runs on DeepSeek-V4.1-Flash (model id deepseek-flash), with reasoning enabled. We may change the model, or offer a choice of models, as better ones appear; the behaviour described on this page applies whichever one is in use.

When you use the assistant, your prompt and the context sent with it (the files you attach, the code in view, the chart or panel you point at, and snapshots of your positions, orders and funds) pass through our servers to the model provider so that it can answer. We do not send your name or email address with it.

DeepSeek processes and stores this data in the People's Republic of China, under its own terms and privacy policy, which say it may use data to improve its models. If you do not want that, do not use the assistant. What we collect and keep is set out in the privacy policy.

Code the assistant writes at your request is yours. We claim no ownership of it and we do not use your workspace, your prompts or your strategies to train models. The model provider's own policy governs what it does with what it receives (see above).

Because models are trained on large bodies of public code, two people asking for the same common thing can receive similar answers, and an answer may resemble existing public code. You are responsible for checking that what you ship or rely on is appropriate for your use.

• Read generated code before running it, especially anything that places an order.
• Backtest, then paper trade, then go live — in that order, with money you can lose.
• Set your risk limits yourself, and do not rely on the assistant to set them for you.
• Do not paste credentials, personal data of other people, or anything confidential into a prompt.
• Keep watching a running strategy. Automation is not supervision.