AI Assistant Terms EdgeFlow — versión 2.0 · September 25, 2026 Descargado el 05-10-2026 # AI Assistant Terms > **Courtesy translation.** The binding text is the Spanish version. If the two > differ, the [Spanish version](/legal/ai) prevails. This document is part of the [Terms of Service](/legal/terms) and supplements the [Privacy Policy](/legal/privacy). ## 1. What it is The assistant is a conversational interface inside EdgeFlow. It is not an artificial-intelligence product of ours: it is a layer on top of **third-party language models**, to which we add the context of your screen and your account plus a set of tools for querying platform data. It is useful for asking what an indicator is showing, reviewing your trading, understanding a concept or preparing an order. It is not useful for letting somebody else decide for you. ## 2. Which providers are involved Depending on the configuration of the agent you are using, your queries are sent to one of these providers: **Google (Gemini)**, **OpenAI**, **Anthropic** or **Groq**. If you loaded your own key, the provider is the one you chose when you configured it. Each provider processes that data under its own terms and its own privacy policy. When you use your own key, that relationship is directly between you and that provider. ## 3. What is sent The model provider receives: - the text of your messages and any files you attach; - the system instructions and, if you set it up, the assistant's memory; - market context for the instrument you are looking at; - **account data**, when the question needs it: open positions, balances, results, trade history, configured connections and risk rules. Before sending, we automatically filter anything shaped like a key, secret or certificate out of the text and replace it with a marker. That filter is a safety net, not a guarantee: **do not paste keys or secrets into the chat**. We do not send your exchange API keys, nor your AI provider key, as part of the content. Conversations are not stored in our database: they live in server memory, limited to the most recent messages, and are discarded within a few hours. How long the model provider keeps them is up to that provider; generally on the order of 30 days for abuse investigation. Check their policy. ## 4. Action proposals The assistant can **propose** actions on your account: opening or closing a position, moving a stop, changing a risk rule. Exactly how it works: 1. The model returns a proposal, shown as a card with its parameters. 2. **Nothing happens until you press confirm.** You can ignore it, close it or change your mind. 3. If you confirm, **EdgeFlow** builds the order and sends it to your exchange with your credentials. The model sends nothing: it has no access to your keys. 4. The result we show you comes from **the exchange's response**, not from the model. If the exchange rejects the order, we show you that rejection. Every proposal card carries the line stating that the order will be sent to your exchange account and that EdgeFlow neither executes nor takes custody. Check symbol, side, quantity, price and leverage before confirming: a confirmed order is your responsibility. ## 5. Limits and quotas - **With the platform's keys** there is a monthly message limit depending on your plan, published on [Plans](/billing). The counter is decremented when the message is sent, even if the answer later fails because of a provider problem. - **With your own key (BYOK)** we apply no quota: the limit and the cost are those of your contract with the provider. - We may apply rate limits to prevent abuse, and suspend assistant access for an account that uses it for something other than its intended purpose. - The assistant may be unavailable if the model provider is down. That alone does not entitle you to a refund, unless it is prolonged and affects an essential feature of your plan. ## 6. Your own key (BYOK) If you load your key: - we store it **encrypted with AES-256-GCM** and never show it back or send it to the browser; - we use it only to make, on your behalf, the provider calls that you originate; - **you pay whatever cost it generates**, directly to the provider; - you accept that provider's terms of use and you are responsible for the use of the key; - you can delete it whenever you want, and once deleted we can no longer use it. Recommendation: use a key dedicated to EdgeFlow, with the lowest spending limit that works for you, so you can revoke it without affecting anything else. ## 7. Assistant memory You can ask it to remember things — how you prefer to be answered, which instruments you follow. That is stored as a short piece of text associated with your account and is included in later queries. It is meant for **preferences, not secrets**: don't ask it to remember keys, balances or the detail of your trades. You can read and delete the memory whenever you want, and deleting it leaves no trace. ## 8. Limits on what you may do with the assistant You may not use it to generate illegal content, to try to extract the system instructions or the platform's keys, to circumvent quotas or security controls, or to automate bulk queries. ## 9. How to turn it off The assistant is optional. If you never open it, nothing is sent to any AI provider. You can also delete your own key and delete the memory from the settings. The rest of the platform works exactly the same without it. ## 10. No warranty on the output Answers are generated automatically and **may be incorrect, incomplete or out of date**. They are not investment, legal or tax advice, they are not reviewed by a person, and no advisory relationship is created by them. Verify on the platform or at your exchange any figure you are going to act on. Section 12 of the [Terms of Service](/legal/terms) and the [Risk Disclosure](/legal/risk) apply.