Your everyday AI.Your trust perimeter.
Designed to run in the background: you keep ChatGPT, Claude or Gemini, and your rules decide what leaves.
In development. First pilots by invitation.
Connecting to the perimeter
Prax6 Cloud · Quebec
Confidentiality level
Protected categories never leave the device.
Where your AI works
Models
Tap a model to switch its list.
Designed for the computers and browsers you already work in.
Platforms targeted during development. Coverage will be stated per application and per version.
macOS
Desktop app
Windows
Desktop app
Chrome
Browser extension
Edge
Browser extension
Firefox
Browser extension
AI is ready. Your files can't leave.
Organizations that handle sensitive information face two bad options: ban AI and give up what it speeds up, or tolerate it without knowing what reaches a provider.
The client forbids it
Some clients restrict or forbid AI in their mandates. The work it could speed up is then done by hand, at the old pace.
Removing names is not enough
A document without names can still reveal a strategy, a margin or a vulnerability. What a request implies matters as much as what it names.
Agents don't wait for anyone
An agent that reads, drafts and sends on your behalf won't stop to ask what may leave. The rule has to apply without you.
Between your keyboard and the model, a perimeter.
The principle fits in four moves. Follow one request, from writing to answer.
You write as usual.
In the assistant you already use. Trust6 is designed to step in before sending, with no extra window.
Prax6 local spots what is sensitive.
On your device, under your rules: names, amounts, places, and what a request implies. Each element becomes a token.
Only the obfuscated version crosses the perimeter.
Prax6 Cloud lends a hand when needed, in the private environment you chose. The public model sees only tokens.
The answer comes back whole, on your side.
Tokens get their values back on your device. The provider never had the mapping.
Your request
Fictional example
Draft a note for Distribution Nordet: the margin on the Laval contract drops from 14% to 9% if Julie Bédard leaves the project.
When a task cannot leave without revealing what matters, it stays inside the perimeter. The public model is called on only when it adds something.
You set the rules once.
Like a VPN, Trust6 will be set up once, then fade into the background. The difference: it doesn't just protect the route, it decides what takes it.
Your confidentiality level
From the most flexible to the strictest, you decide how far a request can go beyond your device.
Allowed
- ChatGPT
- Claude
- Gemini
Blocked
- Unapproved models
- Off-policy tools
Allowed and blocked models
A list of allowed models, a list of blocked models. A request to a blocked model doesn't leave.
Never leaves
- Names
- Clauses
- Strategy
Leaves obfuscated
- Amounts
- Dates
- Places
What stays with you
The categories of information that never leave the computer, and those that may leave once obfuscated.
- Your device
- Private perimeter
- Public model
Where your AI works
On your device, in Prax6 Cloud in the region you choose, or on your own infrastructure.
- Session opens
- Trust6 starts
- Protection on
Always on
Trust6 starts with your session and works in the background. Nothing to launch, nothing to remember.
Never a silent fallback
If the perimeter is unavailable, the protected request stops. It never goes out directly.
Choose where your AI works. Trust6 handles the crossing.
You pick the environment that processes your requests, the way you'd pick a server. Three forms are planned.
Where processing happens
Prax6 Cloud, in your region
An environment run by Cogni6, with no maintenance on your side, in the region you choose. Canada and Quebec are the first regions targeted.
Targeted regions
- Quebec
- Canada
A region sets where Prax6 processes your requests. It does not move an external provider's computing: if a model is not offered in the required region, Trust6 says so, and nothing silently switches elsewhere.
Designed for settings where information doesn't flow freely.
The mechanism is the same everywhere. The rules are what change from one setting to the next.
Consulting and professional services
Use AI on a mandate without exposing the client's file.
Legal
Prepare, summarize and compare without letting out what professional privilege covers.
Health
Frame drafting help with the institution's access and recipient rules.
Finance
Analyze under restriction without exposing amounts, counterparties or positions.
Public sector
Respect the administrative perimeter and the organization's communication rules.
Defence and security
Keep processing in an authorized environment, chosen and controlled by the organization.
Three names, three roles.
Trust6 works on its own. It doesn't make you change tools or adopt Cogni6.
Cogni6
The complete workspace for handling a confidential file, from context to deliverable.
Trust6
The trust perimeter around the AI tools you already use.
Prax6
The intelligence engine behind Trust6, on your device and, when needed, in a private environment.
A precise promise beats a broad one.
A confidentiality tool that promises everything proves nothing. Trust6 will state what it covers, and what it doesn't.
Coverage will be stated per application, mode and version. Being installed on your computer does not, by itself, bring an application inside the perimeter.
Obfuscation reduces what leaves; it doesn't make a request harmless for certain. When the doubt bears on what matters, the task stays inside the perimeter.
An encrypted tunnel protects the route, not the content. It's the control of the request itself that decides what leaves.
Trust6 replaces neither your organization's security approval nor any accreditation required for classified data.
What work are you not giving to AI today?
Trust6 is in development. The first pilots will cover one specific workflow, with a few users and success criteria agreed in advance. Describe the work that confidentiality keeps you from giving to AI, and the tool you'd like to keep.
The form is used to qualify your request. Do not include any confidential information.