Security and privacy
Your AI. Your rules. Your data.
A Node.Bot belongs to you. It computes on your device, and nobody from outside can reach it, not even us. Every action it takes goes through a checkpoint, and it is meant to refuse anything that is not part of its task. Here is what of that runs today.
Principles
What we promise, and from when.
Local models Today
Language, image and search models run on the Node. Cloud models only if you explicitly allow them for a bot.
No access from outside Today
The Node can only be reached on your own network. Access on the go via the hub will come with sign-in and a second factor.
The hub sees no content In progress
The hub knows accounts, Nodes and their online status. Conversations, mail and documents stay on the Node.
Credentials locked Today
No bot can read files with passwords or keys.
Important actions only with approval Today
Lights or music run at once. Unlocking, disarming the alarm or sending a mail only after your yes, with the finished draft to check. An emergency stop halts everything.
Checkpoint for every action In progress
Every tool action of a bot is checked and logged. Today it observes what would be blocked in future. After that, only what is approved is allowed.
Sticks to its task Planned
Today, roles, folders and tools limit what a bot may do. Next, it will also refuse requests that are not part of its job.
Language and courtesy filter Planned
A filter checks tone and content, for children by age group. It observes today and steps in once the legal review is complete.
Operation without internet Planned
A mode in which the Node works only on your own network and sends nothing outside.
Security review before version 1.0 Planned
Before launch we check the system for vulnerabilities. Version 1.0 only comes out without open serious findings.
Support repairs, never reads Planned
Our support may maintain your Node, but not read any content.
Safe updates Today
A backup is made before every update. If something goes wrong, the Node returns to its previous state.
Try it
What happens to this request?
Pick a request and see how the checkpoint decides. The badge shows whether the behaviour already works like this today.
Simplified illustration, nothing is sent.
Request to the bot
Allowed
Reading is allowed. The bot reads the value fresh and answers: “21.4 °C in the children's room, just measured.”
Status: Today
Waiting for your approval
This changes something important or leaves the home. The bot shows you exactly what it wants to do and only acts after your yes.
Status: Today
Blocked
No bot can read files with passwords and keys, however the request is worded.
Status: Today
Declined: not its task
A customer adviser is there for questions about the range. It declines politely and says what it can help with.
Status: Planned
Stay calm, check the tone
The filter recognises abusive language. The bot answers calmly and politely and offers to bring in a person.
Status: Planned
What we do not promise
No system is infallible. We do not promise absolute security or guaranteed child protection. We promise that your data stays with you, that we say openly what runs, and that we close gaps as soon as we know about them.
Questions about security?
In the frequently asked questions we explain which data marnis.ai receives at all.