Prompts (like system instructions) are stored and secured entirely on the backend we've built (and optionally host/manage). Your frontend never holds sensitive prompts or API keys, only the dynamic user inputs are sent to our backend, which then constructs and forwards the complete request to the LLM.