Are my prompts and source code used to train AI models?

Draftbit does not use your code to train AI models, and the code you create in Draftbit belongs to you. Your prompts and the agent’s replies are stored with the thread so you can review the history later. API keys and other credentials you store are encrypted at rest, and traffic is encrypted in transit.

What happens at the model provider depends on which billing path the thread uses. When Draftbit credits pay for a thread, the request goes through Draftbit’s infrastructure to the provider’s API under Draftbit’s accounts. When you connect your own OpenAI, Anthropic, or OpenRouter key, the thread goes straight from the sandbox to that provider under your account. A ChatGPT or Claude subscription works the same way with your subscription credentials, and Anthropic’s policies for Claude subscriptions apply. In every case the provider’s own terms for that account type decide retention and training, so read the terms for the account you connected. The chat and thread history label the provider and billing path for each thread, so you always know which terms apply.

Keep production secrets, customer records, and access tokens out of prompts. Store them as sensitive environment variables, which stay out of the app bundle, and use test data when real records are not needed.

This page is a product explanation, not a certification. If your organization needs a specific compliance standard, data residency, or contract language, ask us for the current terms in writing before you start.