Security
Security that isn't a prompt.
A model can ask for anything. The backend decides what it gets. Here's how Klaie keeps company information with the people who are allowed to see it.
I can't find financial results you have access to. Finance is restricted, but you can ask for access.
Enforced by code, not by asking nicely.
Permissions come first
Access is checked inside every search, in the database. Files you can't open never reach a model on your behalf.
Rules, not prompts
A model can ask to use a tool. Plain, deterministic code decides whether it may.
Every claim has a source
Company and web sources look different, and each citation opens the exact document.
A record of what happened
Sign-ins, sharing changes, tool use and approvals are logged. The content itself isn't.
Companies stay apart
Every table is scoped to one organization with row-level security.
Keys stay on the server
Integration tokens are encrypted at rest and never sent to the browser.
How Klaie uses your data
Your organization's content belongs to your organization. Klaie uses it only to run the product for you.
- We never sell personal information or use it for advertising
- We never use your content to train Klaie's own models
- No advertising or analytics trackers, only the cookies Klaie needs to work
- You can delete conversations and memories, disconnect apps and delete your account yourself, at any time
AI providers
The Klaie 1, Kai 1 and Mol 1 tiers send each request to a model from a third-party AI provider, named in the privacy policy. Some providers offer models on terms that let them learn from what they process. Klaie flags those tiers in the picker, and an administrator can have them switched off for the organization.
Sensitive mode is different. It runs on infrastructure Klaie operates, never sends the conversation to an outside AI provider, and never falls back to one.
Connected apps
Each person connects their own Google or Microsoft account, and Klaie reads with that person's access. New connections ask only for read permissions. Tokens are encrypted at rest, kept on the server, and can be revoked in Klaie or at the provider at any time.
Approvals and calls
Reading and drafting happen on their own. Anything that would send, post or change something outside Klaie waits for a person to approve it. Klaie Call doesn't record calls: audio and screen images are processed live, and a transcript is saved only if you choose.
Questions people ask
Can a clever prompt trick Klaie into revealing restricted files?
No. Prompts can't change permissions. Restricted files are filtered out in the database before a model sees anything, and whether a tool may run is decided by deterministic code, not by the model.
Does Klaie train AI on our data?
Klaie never uses your content to train its own models. Third-party providers behind some tiers have their own terms, which Klaie flags; Sensitive mode never reaches them.
What does the audit log record?
Sign-ins, sharing changes, tool use and approvals, plus opening confidential documents. It records what happened, not the content of your conversations.
Are different companies' data kept apart?
Yes. Every table is scoped to one organization and protected with row-level security in the database.
Keep exploring
Start with one question.
Setup takes about a minute. Bring in files and apps whenever you're ready.
Create a workspaceFree during early access. No card needed.