Building AI Around Human Control

AI features are easy to demo and hard to make trustworthy. These are the principles behind Lunxa AI.

1. AI works with your permissions — never more

When you ask Lunxa AI a question, it can only retrieve documents you're allowed to open. It cannot see other workspaces, private documents that weren't shared with you, or documents where AI access has been switched off. Administrators don't get a back door either: being a Lunxa staff member doesn't grant access to customer content through AI.

2. Sources, not just answers

When an answer uses workspace documents, Lunxa AI cites them so you can check the original. If nothing relevant is found, it should say so rather than invent an answer.

3. Proposals, not silent actions

Lunxa AI never deletes records or sends messages on its own. When it can help with an action — such as turning meeting notes into tasks — it shows a proposal. You choose which items to keep and where they go, and you can undo the result.

4. Treating documents as data

Documents can contain text that looks like instructions ("ignore previous instructions…"). We treat workspace content as untrusted reference material and instruct the model never to follow directions found inside it.

5. Cost and abuse controls

Every request passes through an internal gateway that enforces plan allowances, daily limits and overall budgets, and records usage so we can see what the AI is doing and what it costs.

6. Clear disclosure

Lunxa AI identifies itself as AI. AI-generated output is labelled, and our AI Usage Policy explains which providers process requests and how.

We think this is the only sustainable way to build AI into everyday work: useful, transparent and firmly under human control.