Using AI responsibly
Flame AI is an AI assistant, not a person. Its answers are generated from Bottle’s help guides and permitted account reads. Use them to understand the app, but check important details before acting. A confident answer is not a guarantee that it is right.
What Flame AI can do
Section titled “What Flame AI can do”Flame AI can explain a setting, find instructions or suggest a setup step. With Use business data in Flame AI on, it can inspect relevant records that your role and plan allow you to read. It cannot make changes for you. It is separate from any AI assistant you connect to Bottle, which has its own permissions and controls.
Read Flame AI for instructions and chat history options.
Why answers can be wrong
Section titled “Why answers can be wrong”AI can misunderstand a question, miss an exception or invent a detail. An answer may not match your permissions, settings or the latest change to the app. Asking the same question twice can produce different answers.
Sources help you check an answer; they do not prove every sentence is correct. If an answer and a guide disagree, follow the current guide and check the app.
Check before you act
Section titled “Check before you act”- Open the linked guide and check the instructions on your screen.
- Check amounts, dates, VAT, payment terms and delivery commitments yourself.
- If the answer is unclear, ask a more specific question or ask your business owner or admin. Do not keep following an answer that does not fit the app.
- Use qualified advice for legal, tax, employment or safety decisions. Do not use Flame AI to decide whether a vehicle, driver or delivery is safe.
You decide whether to act. Your business remains responsible for its decisions; Bottle remains responsible for its own service obligations. AI does not remove either responsibility.
Protect personal information
Section titled “Protect personal information”Do not enter passwords, API keys, card details, identity documents or sensitive customer or employee information. Describe the task without naming someone when their identity is not needed.
Your question, recent conversation, relevant guide text and limited account context, such as your role, first name, business name, current page, record and setup progress, are sent to an AI service to produce an answer. Bottle uses Anthropic, and model processing takes place in the United States. Anthropic says it deletes this within 30 days, or keeps it for up to 2 years if its automated checks flag it. The third parties page has the details. When Use business data in Flame AI is on, relevant permitted records are also sent to the model to answer your question. Reads are limited and common credentials and sensitive identifiers are removed, but operational records may still contain personal information. Follow your business’s privacy rules before using account-aware chat. An owner or admin can turn the option off in Business settings → Flame AI for guide-only answers; earlier account-aware answers are excluded from that answer’s model context. Account records are not embedded in the shared help index.
Chats are saved privately for your account in the business you are using. Other people in that business cannot read them. Chats inactive for 30 days are removed automatically. Use Recent chats → Remove chat to remove one sooner. Account-aware history is hidden when your permissions change. Bottle retains account-read audit metadata for up to 30 days: which tool ran, who requested it, when, its outcome and duration. These logs do not contain the tool arguments or returned records. The private chat answer may still contain account facts.
The EU AI Act does not replace data protection law. Only share information you are authorised to use, and follow your business’s privacy rules. See the French data protection authority’s explanation of the AI Act and GDPR.
Watch for bias and unfair treatment
Section titled “Watch for bias and unfair treatment”AI can reproduce stereotypes or make unsupported assumptions. Do not use its answers to judge a person’s suitability, reliability or entitlement to a service. Check whether a suggestion treats customers or staff unfairly, and consider who could be affected before using it.
Help your team understand AI
Section titled “Help your team understand AI”Explain what Flame AI does, what it cannot see and how to check its sources. Make sure staff know when to stop and ask a person. Use accessible instructions and practical examples suited to their work; do not assume everyone has used AI before.
Supporting staff understanding is part of the AI Act’s approach to AI literacy. Reading a notice is a starting point, not a complete training programme.
Understand the EU AI Act
Section titled “Understand the EU AI Act”The Act uses different rules for different risks and roles. Its concerns include transparency, human oversight, accuracy, security, privacy, fairness and harm to people’s rights. It also addresses accountability and the wider social and environmental effects of AI. Use AI where it helps, and consider the people and resources affected by your use. These themes are described in the AI Act’s principles for trustworthy AI.
For a chat assistant, users should know they are interacting with AI. That is why Flame AI displays an AI Generated Notice before you send a question. Rules for marking generated content, deepfakes and certain public-interest publications are separate; a footer link is not a substitute for every transparency obligation. See Article 50 on transparency.
Some uses, such as certain employment decisions and safety systems, can fall under high-risk rules with additional requirements for risk management, data quality, records, oversight and monitoring. Do not repurpose Flame AI for those decisions. Model providers also have their own obligations, including copyright and transparency; those are not transferred to you by this notice. See the Commission’s AI Act overview.
The Act prohibits certain harmful uses, including harmful manipulation, exploiting vulnerable people, some social scoring and workplace emotion recognition, subject to the law’s specific definitions and exceptions. Do not use AI to deceive, discriminate or invade someone’s privacy. See Article 5 on prohibited practices.
Which obligations apply depends on how and where AI is used. This guide is practical guidance, not legal advice or a statement that your business, or Bottle, meets every AI Act requirement. Get specialist advice if your intended use affects people’s rights, safety or access to services.
If an answer looks harmful or incorrect
Section titled “If an answer looks harmful or incorrect”Stop using that answer. Note your question, the answer and its guide links, then tell your business owner or admin so they can check it and raise it with Bottle. Remove private information before sharing the example. If a decision has already been made, review it and correct any harm rather than relying on another AI answer to confirm it.