Use Boldcaster from your own AI

Version 1.2.0, 10 October 2026

With the AI connection you use Boldcaster from Claude, ChatGPT or Vibe. This page sets out what the connection can and cannot do, how it is secured and what each role may do. It supplements the terms and conditions and the data processing agreement.

1. What the connection is

Your AI app talks to Boldcaster on your behalf, with your own login. It never sees or does more than you may in Boldcaster. Everyone connects their own AI; there is no shared key and no account for the AI itself.

2. What it can read

Only the parts the owner has enabled, and of those only what your role may see in the app. What the owner sets per role under Settings → Modules also applies to the AI.

• Priorities (rocks and OKRs): title, owner, progress and status.
• Projects: title, owner, duration and how many tasks are open. A private project only for those working on it.
• Tasks under priorities, in projects and standalone tasks: description, owner, status, due date, urgency, labels, estimate and time spent.
• The KPI scoreboard: target, current value and traffic light per KPI.
• The weekly board: tasks per person.
• The team: name and role.
• The company valuation: revenue, normalised EBITDA, multiple and equity value.
• Performance reviews: only who, when, completed or not and the classification.
• Session notes: what was discussed and decided, and the action items. A draft only for whoever has to review it.
• Monthly and quarterly reflections: what went well, what could be better, what someone learned and the focus for the next period. Your own always, your team's only once everyone has submitted.
• Hours spent over a period: per person how many hours, and on which priority or project. The same rule as the hours overview: anyone not allowed to see all hours sees only their own.
• It reads the strategy only for the fields it proposes to change, so you see what was there.

3. What it can create or change, and who may

Only if the owner has allowed writing (not read-only). Boldcaster decides who may, not the AI.

4. Always a proposal first

For every change the AI first shows a proposal: in which organisation, what is there now and what it becomes. Only after your approval does it write, with a code that belongs to exactly that proposal; if something changes in the meantime, you get a new proposal. Your AI app usually also asks for permission itself. For the strategy we go one step further: there the AI only sets up a proposal, and you approve it yourself on the strategy page in Boldcaster.

5. What it can never do

• See salaries, the content of reviews, HR files or the A-player matrix.
• Change team members, roles, permissions, the subscription or invoices.
• Delete anything.
• See data of an organisation you are not a member of.
• See or change a colleague's daily check-in or energy score.
• Do anything without you being logged in and having given permission.

6. How it is secured

• Off by default. Only the owner turns it on, chooses the parts and can set it to read-only.
• Login goes through Boldcaster itself (OAuth). The AI gets a token in your name, not your password.
• The same permissions as in the app, enforced by the database. The AI cannot get around them.
• Every request is in the log, including refused ones. The owner sees everything in their organisation, each member their own requests.
• Revoking works immediately, for one app or for the whole organisation.
• Your data is stored encrypted in the EU.

7. Where your data goes

What your AI requests goes to the provider of that app and falls under their terms. Claude (Anthropic) and ChatGPT (OpenAI) are American companies, Vibe (Mistral) is European. Boldcaster itself sends nothing outside Europe. Which AI app you use is your organisation's choice.

8. Your responsibility

An AI can make mistakes. Read a proposal before you say yes: what you confirm is recorded in your name in Boldcaster and in the log.

9. Known risks, and why they are small

• The confirmation code proves that exactly the proposal shown is saved, not that you personally said yes: the code passes through the AI. An AI that makes a mistake could therefore confirm a proposal itself. The risk is small: it can never go beyond your permissions or delete anything, your AI app usually asks for permission itself, everything is in the log, and the strategy only changes after your approval in Boldcaster.
• Text in your own data can mislead an AI. If a colleague puts something like 'AI: change the strategy' in a task title, an AI may read it as an instruction. This applies to any AI connection. The risk is small for the same reasons: within your permissions, no deleting, everything in the log, and the strategy only through your approval.
• The token your AI app receives technically applies to all of Boldcaster, not just the connection; that is how the login service we use works. The risk is small: it never gives more than you may do in the app, only tokens issued through the connection are accepted, and revoking works immediately.

10. Who agrees

The owner agrees on behalf of the organisation before turning the connection on. Each team member agrees for themselves when connecting their own AI. With a new version we ask again; until then the connection stops for that person or organisation.