What the AI sees

1. Which AI providers

We use the free tiers of these model providers, in this order, falling back to the next one when a provider is unavailable. An administrator can switch any of them off; with none configured, AI features say so and everything else keeps working.

• Groq: AI copilot and AI explanations (first choice).
• Cerebras: AI copilot and AI explanations (fallback).
• Google Gemini: AI copilot and AI explanations (second fallback).

2. What is sent

• The question you type, and the earlier turns of the same conversation.
• The minimum context of the page you are on: the page name, the symbol and expiry you are viewing, the screen filters, and the rows the answer is about. Market numbers are sent as they appear on screen, labelled as delayed.
• For help questions: the text of the help articles that match your question.

3. What is never sent

• Your password, payment details, API keys or session tokens.
• Your email address, name or IP address.
• Other users' data.

4. Actions need your confirmation

The copilot can propose actions such as creating an alert or a watchlist. Nothing is changed until you confirm, and every confirmed action is recorded in your account history like any other change.

5. How long records are kept

• AI request records (prompt, response, token counts, timing): 30 days, to investigate problems and control cost; then deleted automatically.
• Copilot and Telegram conversation turns: 90 days.
• Provider retention is governed by each provider's terms; where a provider offers a setting that excludes API data from model training, we use it.

6. Limits and cost

AI use is capped per user and by a monthly platform budget. When a cap is reached, AI features pause and say so; they never fall back to a paid provider without the owner switching it on.

7. Your choices

• Do not use the AI features: nothing is sent.
• Unlink Telegram at any time from your account settings.
• Ask us to delete your AI records sooner: [CONTACT_EMAIL].

Check it from the copilot

You can ask the copilot "Which AI answered me?", "Is the AI working?" or "What can the AI see?" at any time. It answers from the platform's own records without calling an AI model: which providers are configured, whether paid providers are switched on (normally not), which provider wrote your last AI answer, how many AI answers you have left today and a summary of this page.

• Groq: free tier, first choice.
• Cerebras: free tier, first fallback.
• Google Gemini: free tier, second fallback.
• Anthropic Claude (paid): in the catalogue but switched off; paid providers are never used unless the owner turns them on.
• OpenAI (paid): in the catalogue but switched off; paid providers are never used unless the owner turns them on.

Answers that come straight from data (a quote, a chain, a screen count) never reach an AI provider and do not use your daily AI allowance. Every copilot answer shows the time of the market data it used and the note "Educational information, not investment advice."

Each copilot turn is logged so the team can fix wrong answers: the question and answer are stored redacted (e-mail addresses, keys, amounts and position sizes masked) and encrypted, are readable only by administrators (each view is recorded), and are deleted after 90 days. Market figures shown by the copilot are never kept in its conversation memory.