Security and responsible AI
Useful agents, with people holding the authority
These are the rules the product is built around, not a policy written afterwards.
Your workspace is separate
Every document, task and result belongs to one workspace. No agent can read another organisation's material.
Least privilege by default
An agent receives only the tools and information its job requires, and nothing more.
Humans approve high-impact actions
Payments, deletions, public posting and contacting people outside your workspace always need a named approver.
Draft-only until you decide otherwise
New agents start producing drafts. More autonomy is unlocked only by an administrator.
Honest about uncertainty
Outputs are labelled as AI generated, cite the material used and flag conclusions the data does not support.
Everything is traceable
Who assigned the work, which knowledge was used, who approved it and when — kept in an audit record.
Secrets stay on the server
Provider keys and access tokens are never sent to your browser or shared between workspaces.
Support access is limited
Support staff cannot read your documents by default. Elevated access needs a reason, a time limit and a log entry.