Chat With Your Data is powered by a large language model (LLM): currently Anthropic, which is hosted on Amazon Web Services (AWS Bedrock) within Clue's own secure AWS environment.
Rather than letting the AI roam freely, we've built a strict set of "tools" that the LLM uses to look up specific things from your tracked data. These tools have been coded by Clue engineers, with your data privacy and security in mind. The AI is allowed to understand your question in natural language, then only use the approved tools to fetch the relevant information and write an answer.
These defined tools work as a guardrail to limit what the AI can do. The chat can only ever work with your tracking data and the conversation data you enter into the chat yourself, nothing else. Your data is never used to train this LLM, or any other AI model. You can read more about data processing in this article.
We’ve also created safety filters, so that before your question even reaches the AI, it passes through a check to see if it contains high-risk content. These filters have been designed by our Science Team to ensure accuracy and safety. Questions in sensitive areas (such as diagnosis, treatment, pregnancy, or contraception) are caught by this filter and answered with a fixed, pre-approved response built by Clue. The AI is never involved in these cases.