AI Disclosure

Alltra uses AI in a small number of specific places, and nowhere else:

• Alltra Intelligence — the chat surface, where you ask questions about your own trading and it answers by querying your records.
• Actions — prepared analyses you can run against your data from the Intelligence surface.
• Extraction — reading structure out of things you have already written, such as a journal entry, so it can be summarised or grouped.
• Memory — durable notes the assistant keeps so you do not have to repeat context in every conversation.

Nothing else in Alltra is AI-driven. Your trades, your statistics, your calendar and your reports are calculated arithmetically from the executions your broker reported. No figure anywhere in the product is produced by a model.

When you ask Intelligence a question, your message is sent to a third-party model provider. The assistant then answers by calling tools that query your own records, and what those tools return is sent to the provider as well, so the model can read it.

In practice that means the following can reach a model when a question calls for it:

• The messages you type, and any file you attach to a conversation.
• Trades returned by a query — including symbols, direction, size, result and any notes you attached to them.
• Journal entries and their titles, where the question is about what you wrote.
• Strategy names, tags and account names.
• Memories the assistant has saved, if memory is enabled.

Data is sent per question, in response to what you asked. Alltra does not send your account to a provider in bulk, and a query that does not need your journal does not send it. It is worth being clear about the consequence anyway: if you ask a question about your journal, your journal is what gets sent.

Your password is never sent. Neither are your broker credentials, which are held for the connection itself and are not part of anything the assistant can query.

Alltra does not train or host its own models. Model processing is carried out by a third-party provider on our behalf.

Model processing is carried out by DeepSeek, whose models answer the assistant’s questions, draft journal entries, and read the images you attach. Where a document provider is configured, a PDF statement you attach is read by Google, through its Gemini models; where one is not, Alltra refuses the file rather than sending it elsewhere. Alltra is built to support more than one provider, so this may change; when it does, this page is updated, and the current providers are always the ones named here.

Alltra does not use your trading data, journal entries or conversations to train models of our own. We do not have any.

Whether the third-party provider may use data we send it for training its models is governed by our agreement with that provider rather than by anything in this product, and it is set out in our Privacy Policy alongside the rest of our processor arrangements. If the answer to that question matters to you before you use the assistant — and it is a reasonable thing to want in writing — ask us and we will give you the current position directly.

Most of what the assistant does is read. The majority of its tools are queries against records you already have.

It can also write. It can create and edit trades, accounts, account adjustments, platforms, transactions, budgets, goals, journal entries, emotion logs and memories, and it can delete most of those. What it cannot do is any of that quietly. Every tool that is not a read stops before it runs and comes back to you as a confirmation card naming the action and showing the exact values it would use:

• Nothing is written until you act on that card. The card’s wording and the figures on it are composed by our servers from the request the model actually made — they are not written by the model — so it cannot describe an action as something milder than what it would do.
• You can edit the values on the card before confirming it. Which fields are editable is decided by our servers from the tool’s own definition, not by the model and not by your browser: an account the request already resolved cannot be re-pointed somewhere else by editing the card.
• You can say “Always allow” for a tool you would rather not confirm every time. That permission is stored against your account, is listed in Settings › Intelligence, and can be revoked there at any moment.
• A standing permission can never cover a destructive tool. Deleting a trade, a journal entry, an emotion log, a goal, a budget category, an adjustment or a transaction — and archiving a platform — always needs a fresh confirmation. That restriction is applied when the permission is granted and checked again every time it is used, so a tool that becomes destructive later loses every permission already held over it.
• An action that ran under a standing permission is recorded exactly as a confirmed one is, and appears the same way in your activity log. Granting one removes the button, not the record.

Two limits are worth stating plainly, because they are properties of the product rather than promises about the assistant. It cannot move money: Alltra is not connected to any payment rail, and a Portfolio transaction is a record you keep rather than a transfer anyone makes. And it cannot edit your statistics directly — every figure in Tracker is derived from your trades, so the only way it can change one is by changing a trade, which is a confirmable action like any other.

Memory is on by default and can be turned off in Settings. Some things about how it behaves are worth stating precisely, because "delete" means different things in different products:

• Turning memory off stops the assistant reading your memories. It does not delete them — deletion has its own control, so that switching the feature off and on again does not silently destroy anything.
• Deleting a memory is a hard delete. The record is removed, not hidden behind a flag.
• A deleted memory stays deleted. To stop the same fact being extracted again from the same source, Alltra keeps a one-way fingerprint of the deleted text — not the text itself — so the record of what you asked to forget is not a copy of the thing you asked to forget. Those fingerprints are removed after 180 days.
• Conversations you delete are removed from your account and retained in backup for a limited period before being purged, in line with the retention described in our Privacy Policy.

AI output can be incorrect, incomplete or confidently wrong, and this one is no exception. The assistant is a way of asking questions about your own records quickly; it is not a source of truth about them. Where an answer matters, the underlying page is there and it is the one to check.

Nothing the assistant produces is financial, investment, tax or trading advice, and it is not a recommendation to enter or exit any position. This is the same position that applies to the rest of Alltra and is set out in full in our Disclaimer. Decisions about your money remain yours.

This page describes how the assistant handles your data. It does not replace or vary anything in our other documents: the Terms of Service govern your use of Alltra, the Privacy Policy governs how we handle personal data generally, and the Disclaimer sets out the limits of what Alltra is. Where this page and one of those documents appear to disagree, the other document is the one that governs.