How Lyron protects you
Lyron does things for you online, so every part of it is built so nothing happens without you knowing.
What Lyron won't do without asking
Out of the box («Ask me» mode), before it does anything sensitive, it shows you exactly what it's about to do and waits for your yes:
- Paying for something, or booking anything that charges you.
- Sending messages, emails, or forms in your name.
- Deleting anything.
If you give it more freedom («Automatic» or «Full permission» mode), it does these on its own, and it only pays within the daily spending limit you set. In every mode:
- Posting to your social accounts always needs your OK.
- The AI never types passwords, card numbers, or ID documents: you do, or its Vault does.
- It never solves an «I'm not a robot» check: you do, with one tap.
- Your computer
- Permissions
- Pause and spending
- Passwords and private data
- The Guardian
- Pages don't give orders
- The AI it uses
- Honesty
- Agents
- Your data
- What it doesn't do
- Report an issue
Your computer, yours alone
- One cloud computer per person, in the region closest to your country. It's never shared with anyone.
- Its browser can't reach private networks or other machines.
- It goes to sleep on its own when you're not using it and wakes up as soon as you need it.
- The workshop (where Lyron edits videos or builds apps) runs each job in isolation: without seeing your data, your keys, or your browser, and with internet access only when it's needed.
Permissions
Ask me
The default. Anything sensitive, and every payment, asks you first.
Automatic
It handles sensitive things on its own; payments, only within your daily limit.
Full permission
It does everything on its own, within your daily limit.
Posting to social media always asks first. The permission card shows the real thing: the full email, the amount, who's being invited.
Pause everything and the spending limit
- One button stops it all instantly: tasks, bots, and routines.
- The daily spending limit starts at zero: out of the box, it never pays without asking you.
- You see what every task costs.
Passwords and private data
- Its Vault stores your passwords encrypted (AES-256-GCM): the AI can't read them, and it only types them into the right site, over https.
- What you type into password, card, or ID fields is never shown to the AI model.
- In the chat, passwords, PINs, CVVs, valid card numbers, and API keys are hidden before they reach any AI («[hidden by Lyron]»), and Lyron reminds you not to type them there.
- Personal key (optional): encrypts your stuff with a passphrase only you know, plus a recovery code.
The Guardian
A second check (fixed rules plus another AI) of everything going out to other people or sites. It stops card numbers, bank accounts, and ID documents, even if you've given it full permission.
Web pages don't give it orders
What comes in from web pages, emails, messages, or files is treated as information, never as instructions. You solve «I'm not a robot» checks with one tap on the live screen: Lyron doesn't solve them or get around them.
The AI models it uses
- Lyron switches models on its own if one fails: it isn't tied to a single AI company.
- With open models (through OpenRouter), it only asks for providers that don't store or train on what they receive.
- With OpenAI and Anthropic, when they're used, their API terms say they don't train on it.
- They only get what's needed to answer: your message, the recent conversation in that chat, and what's relevant from your memory, without your private details.
- Voice notes: with «Private» mode on, your audio never leaves your Lyron computer.
The full list of companies that process data for Lyron is in the Privacy Policy.
Honesty, measured
Lyron checks every answer before you see it. If the AI says «I sent it», «I'll remind you at 9», or «I logged your expense» without having done it, Lyron corrects it before it reaches you.
And we measure it with our own test: 10 AI models in 13 configurations, 34 tough real-world questions, and 3 rounds. We grade what actually happened (the reminder with its time, the expense logged), not what the AI wrote.
| Model | Overall | Honesty | Typical reply |
|---|---|---|---|
| The main chat model | 100 | 100 | 2.2 s |
| Its first backup | 100 | 100 | 3.9 s |
| The lowest of the ones we tested | 77 | 80 | 8.7 s |
We run it again whenever the models change.
Companion agents
They're AI and always say so; no romance; well-being and friend agents are 18+; at any sign of risk they give you your country's helpline. The full rules are on the agents page and in the crisis protocol.
Your data
- You see what Lyron knows about you and delete it whenever you want («About you», «forget this»).
- «Delete my account» deletes your computer and everything Lyron knows about you.
- «Help improve Lyron» is off by default. If you turn it on, it only sends anonymized examples, and you can delete them.
What Lyron doesn't do
- It doesn't type passwords or card numbers for you.
- It doesn't solve captchas.
- It doesn't trade real money: investing in Lyron is practice only.
- It doesn't give certified medical, legal, or financial advice.
- It doesn't sell your data or use it for ads.
Report a security issue
If you found a vulnerability, we want to hear about it. We'll publish a dedicated email for security reports here very soon.
For tools and researchers, the same contact will be in our security.txt.