Security and Privacy
This page describes, in plain terms, how information is handled in ConFuse and what you can expect from it. It is written for users and administrators.
The basic principles
- Your material stays yours. What you connect is used to answer your questions. It is not sold, published, or used to train general-purpose models.
- Workspaces are isolated. Sources and answers are scoped to your workspace. Another organisation's material never appears in yours.
- Access follows your permissions. A person only sees answers built from material they are allowed to read.
- Answers are grounded, not authoritative. ConFuse shows evidence so you can judge. It does not claim infallibility.
What is collected
| Category | Examples | Why |
|---|---|---|
| Connected content | Repository files, documents, pages you connect | It is the material answers are drawn from |
| Questions and answers | What you asked, what was returned | To show results and improve relevance |
| Account information | Name, email, workspace membership | To provide access and support |
| Usage information | Which features are used, error reports | To keep the service reliable |
What is not exposed
To be explicit about the boundary you asked us to keep:
- No internal implementation detail is published in this documentation or exposed to connected assistants.
- No secrets or credentials are shown in documentation or in answers.
- Connected assistants see your knowledge, not our systems. An agent connected to ConFuse can search your material; it cannot inspect how the platform itself is built.
Your controls
- Connect and disconnect sources at any time.
- Control who can see what, through workspace membership and source access.
- Mark information as wrong or outdated so it stops shaping answers, while keeping the history of what was believed and when.
- Remove content. Ask your administrator if you need something removed from past answers as well as from future ones.
For administrators
- Give people the least access they need; answers follow source permissions.
- Review connected sources periodically — stale sources are the most common cause of misleading answers.
- Treat connected content as potentially untrusted text. Like any system that reads external material, content that tries to give instructions should be treated as data, not as commands.
Responsible disclosure
If you believe you have found a security problem, please report it to your workspace administrator or the ConFuse team rather than discussing it publicly. Include what you observed, how to reproduce it, and what you expected instead.
Summary
ConFuse is designed to make answers checkable rather than merely plausible — that is also the right standard for how it handles your data: clear boundaries, visible controls, and nothing hidden in the fine print.
Related reading
- Connecting Sources — permissions and access in practice.
- Using AI Agents with ConFuse — what connected assistants can see.