AI Trust and Ethics

How our AI works, in plain words

Fuela is the AI inside our OS. Here is exactly which models we use, what the AI can see, who stays in control, and what to do if something goes wrong. Fully transparent, for anyone to read.

Last updated October 19, 2026. Read with our Privacy Policy, Terms of Service and EULA.

The AI we use, named

Fuela, our AI, runs on Anthropic's Claude models by default. Claude also handles translation.
Some features can use other providers: OpenAI (speech to text as a backup, and image making), Google Gemini (an optional brain a team can pick), Deepgram (speech to text for calls and voice notes) and Eleven Labs (Fuela's speaking voice).
If we add or change a provider, we update this page and the changelog below.

Your data does not train AI models

We do not train AI models on your content, and we never sell it.
We send your content to these providers through their business APIs, whose published terms say they do not train their models on that data by default. For speech to text with Deepgram, we switch off their model improvement program on every request.
We never train a publicly available AI model on your content. Our privacy policy says the same.

The AI only sees what you can see

When you ask Fuela something, she reads only what you are already allowed to see: your rooms, your tasks, your docs. Nothing past that.
Every doc, deck and room has a sharing level: private, team, brand or public. The AI respects the same levels people do.
A family or a company can run private first, so nothing leaves their own space.

People stay in control

Fuela suggests and acts only inside the permissions of the person she is working for.
Anything that reaches the outside world waits for a person: sending messages to people outside the team, spending money and deleting data.
What Fuela does shows up in the chat, task or doc where she did it, so you can see what happened and who asked.

AI is always labeled

Every bot carries an AI badge, in chat, in replies and in lists, so you always know when you are talking to AI.
Bots have their own colors and rings on their avatars, so they never pass as people.

Kids are protected

Young builders never chat with adults. Adults only see that a young builder is online.
A parent or guardian applies, gives consent and can withdraw it at any time. Parents and approved staff moderate.
How it works, step by step, is on the Young Builders page.

Honest about what it knows

Fuela answers from your own records and says where an answer came from.
When she does not know, she says so instead of guessing.
AI can still get things wrong. Check anything important, and tell us when it misses, so we can fix it.

Halal principles for Muslim mode and Deen Fuel

Inside Muslim mode and Deen Fuel, Fuela does not promote what Islam prohibits: alcohol, pork, drugs, gambling, interest-based products and explicit content.
Answers on matters of faith are meant to rest on the Quran and authentic hadith, and Fuela says when a question is a matter of scholarly difference and should go to a scholar.
Honest status: these are our commitments and the direction we are building toward. No scholar has reviewed or certified them yet, so we do not claim a halal certification.
Next step: a Sharia advisory board of two or three recognized scholars who review and sign off, the realistic form of halal assurance for software. Until it exists and publishes its sign-off here, nothing on this page is a fatwa or a certification.
Every other faith, and no faith, is equally welcome across the OS. Muslim mode is a choice a person turns on for themselves, and it changes nothing for anyone else.

Report a problem

Use the Help Desk inside the OS, or ask Fuela to report it. A person reads every report.
Not a member? Write to us from the contact page.

Delete your data

You can delete your own messages, docs and uploads, and ask us to delete a recording or transcript.
To delete your whole account, ask through the Help Desk. Our privacy policy explains how long each kind of data is kept.

Changelog

October 19, 2026: Added halal principles for Muslim mode and Deen Fuel, with the honest status: commitments, no scholar sign-off yet (#8829).
October 16, 2026: First published: models, training, access, control, labels, kids, accuracy, reporting and deletion.

Fuel Cart