AI data and privacy
What data ProjectBrain's AI features send to model providers, what stays hidden, and the controls admins have.
When you use an AI feature, ProjectBrain sends the text needed for that one request to an AI model, then shows you the answer. This page explains what that text includes and how to limit it.
What is sent
Section titled “What is sent”Only what the feature needs. For example:
- Drafting a brief sends the template, the project details and your answers to its questions.
- Drafting a blueprint sends a summary of the work it is based on.
- A standup suggestion is built inside ProjectBrain from your own recent work. It does not need a model to list what you did.
ProjectBrain uses third-party model providers, such as Anthropic and OpenAI, to run these requests. Which provider handles a feature can change as models improve.
What AI does not see
Section titled “What AI does not see”- Anything you cannot see. AI features run as you, so private projects, ethical walls and 1:1 standups you are not part of stay out of reach.
- Pricing, when it reaches a connected assistant. Rates, costs and SOW pricing tables are hidden from a connected assistant unless the person who connected it turns on Share pricing with this AI for that connection.
- When a document draft draws on an earlier SOW, the pricing sections of that SOW are left out.
Connected assistants
Section titled “Connected assistants”When you connect Claude, ChatGPT or Codex, that assistant reads ProjectBrain data through your connection and sends it to its own provider, under your account with that provider. Your agreement with that provider decides how they handle it.
A connected assistant never sees email addresses of your teammates. It sees display names and roles.
Controls for admins
Section titled “Controls for admins”People with the mcp.admin permission manage these at Settings > MCP Policy:
- Providers: allow every AI provider, or only the ones you pick, or block the ones you pick.
- User access: let everyone in the organization connect an assistant, or only chosen people.
- Limits & autonomy: a daily request cap for the organization, and whether bots need a person to approve their actions.
- Emergency kill-switch: blocks every connected assistant in the organization at once, from the next request.
Each person can see what their own connections did at Settings > MCP Activity.