Vellix Business OS
Security and trust: how people stay in control.
Most security pages are a wall of badges. This one describes how the controls actually work, and further down there is a plain ledger of what is built, what is tested, what is planned, and what nobody has certified.
The control model
Twelve controls, one rule: nothing important happens without permission.
People and AI steps alike get only the access their job requires, and nothing more. That is the default here, not something you have to ask for.
Every connection to your systems is limited to what it needs, and you can switch it off. We list a connection publicly only once it is built and tested. None are advertised today.
Standard Edition does not allow patient health information at all. The planned Healthcare Edition is switched on separately, for healthcare's stricter rules. See where it stands today.
Every customer's information and jobs are kept apart from every other customer's, by design.
Anything out of date, expired, contradictory or shaky is flagged, rather than treated as fact.
Designed to work with different AI models and services, including ones running on your own hardware where that is allowed. Nothing about the design ties you to a single supplier.
Anything your Blueprint marks as sensitive is checked against your rules and then waits for a person to say yes. Before it happens, not after. In the tested foundation the approving step is kept separate from the AI step being approved.
Improvements come from results someone has checked, and each change is put to a person. Nothing shifts quietly in the background.
Important actions are recorded: who or what did it, when, and under which rule. Knowing who did it is the difference between trusting the system and hoping.
What AI ran, for which job, and what it cost. Visible, measured, and tied to the work it did.
Spending limits and warning levels that you set. The point is to stop surprises, not to stop the work.
A change is suggested, reviewed, approved, dated, and can be undone. The Blueprint cannot quietly rewrite itself or anyone's access.
The honest ledger
Built, planned, or just an idea? We say which.
Most vendor security pages blur these together. Here is where each of our claims actually stands, as of 2026-09-11.
| Claim | Status | What that means |
|---|---|---|
| Keeping each customer's information separate, access by role, sign-off stops, the record of what happened, AI cost tracking, and dated Blueprint changes | In private testing | Internally tested, and running only in our private test environment, on made-up data. Our tests have found defects during development, and accepted corrections are re-tested before a release is frozen. No customer is using the Business OS, and this is not an independent security audit. |
| Customers being able to sign in | Not available | Not open to customers, on purpose. The only accounts are our own test accounts, and no customer can get in today. That is a decision, not something we forgot. |
| Named connections to other software | Not available | We are not advertising any. A connection is named here only once it is built and tested. |
| Patient health information in Standard Edition | Prohibited | Protected health information may not go through Standard Edition. Full stop. |
| Healthcare Edition's controls | Planned | A future version, switched on separately. Not available, and we make no compliance claim for it. |
| HIPAA compliance | Not available | Not claimed — no Vellix product claims HIPAA compliance. What HIPAA requires of "business associates" is defined by the U.S. Department of Health and Human Services — see hhs.gov's HIPAA professional guidance. What we do promise today is narrower and factual: DenialIQ only handles patient health information once a Business Associate Agreement is signed, and pilots start on data with the patient details stripped out, or on made-up data. |
| SOC 2, or any other outside certification | Not available | None yet. Nobody independent has audited us. We would rather say that plainly than hint otherwise with a badge-shaped graphic. |
Things we will not say, because nobody can honestly say them: that your data cannot be stolen, that any platform is breach-proof, that everything is encrypted everywhere, or that putting something on a particular cloud automatically makes it compliant.
People, not just policies
It works from what you write down. It does not watch your staff.
Business OS is designed to work from your Business Blueprint: the business information, rules, system map and approvals your company chooses to put in it. Records in your connected systems are used only for work you approved. Each step is designed to see only what that step needs, and a person reviews changes. It is not an employee-monitoring tool, it is not designed to watch people on the quiet, and by design the Blueprint is not able to change itself or anyone's access. If we cannot explain a feature to the team using it, it does not belong in the product.
Two pages carry the detail behind this one: what a Business Blueprint contains, which is the written record all of the above works from, and what a human-in-the-loop AI system actually looks like, which explains why an approval has to be a separate step from the work being approved.
Security questions, including the hard ones: contact@vellix.ai. Please don't include sensitive or confidential information in an email — we'll set up an appropriate channel if the conversation needs one.
Bring us your security requirements.
Design partners help set the bar. Tell us what your business, or your industry, would need to see before trusting AI with real work.