1. What the service does
BotStacks helps customers configure assistants using instructions and connected knowledge sources. Retrieval, generated answers and testing can produce errors. A cited source does not guarantee an answer is accurate. Illustrative demonstrations are not measured customer performance.
2. Inform people before interaction
Deployment requirement: identify the assistant as AI in a clear, accessible notice at or before the first interaction, link the deploying business’s privacy notice, and explain how to contact a person. Do not hide the disclosure in a footer or represent the assistant as a human employee. BotStacks must assess its own provider obligations; customer instructions do not transfer those obligations away.
Suggested end-user notice: You are chatting with an AI assistant. Answers may be inaccurate. Avoid sharing sensitive information. See this business’s privacy notice or contact its team for help. Customer-specific links and the actual escalation method must be supplied before deployment.
3. Tools and persistent memory
Where enabled, tool integrations can read information or perform actions using granted permissions. Limit access to the intended workspace and purpose, obtain authorization for consequential actions, and make failures visible. Do not authorize payments, deletion or external communications merely by accepting an AI response.
Where persistent memory is enabled, explain what is retained, the purpose, who can access it, and how correction or deletion is requested. Memory must follow the customer’s instructions, retention schedule and applicable consent requirements. This document does not claim that Mem0 or any particular integration is active for every customer.
4. Restricted use and oversight
The Acceptable Use Policy prohibits unlawful conduct and sole reliance on AI for consequential decisions. Human review must be meaningful and does not legalize a prohibited use. Regulated deployments need an approved use-case assessment and any required contracts before activation.
5. Training, provenance and reports
Model-provider training and retention settings depend on the applicable provider and configuration. Content-grounded retrieval and workspace configuration are not the same as training a general-purpose model. Any additional no-training commitment must be expressly documented in the applicable order or DPA.
For deployments within EU AI Act scope, assess both interaction transparency and applicable machine-readable marking of generated outputs. A visible AI label alone does not establish full Article 50 compliance. Report harmful or misleading behavior through [email protected], including only the minimum information needed to investigate.