Your personal tech support, on call 24/7.

An AI helper you install on your computer. She troubleshoots problems, can notice crashes, look things up online, and run tools when needed. A harness decides what is allowed, what needs your OK, and what is blocked.

On your computer. The app, the safety rules, shell commands, files, crash list, and the built-in browser all live there. Talking to the AI and searching the web go through your OpenRouter account. Our servers host this website and the download page, not your chats.

What she needs to help you in that conversation goes to OpenRouter and the AI models behind it: your messages, files or screenshots you attach, command output, pages she reads, and crash details if you ask her to investigate one. She is not quietly copying your whole disk or sending every crash file to us. The app may check our site for updates. Your chat does not go through a Lizzandra chat server today.

No. Before a command runs, the app itself checks allow, ask first, or block. The AI cannot rewrite or turn those rules off. Asking her to be careful helps, but it is not what actually stops a bad command. You still choose the profile, can edit rules, and approve or deny what the harness asks you about.

Yes. Any tool that can fix your PC can also make mistakes or do harm if you loosen the rules, approve something you do not understand, or run a riskier profile. AI can misread a situation and propose a bad command. The harness blocks many of those, but it is not magic, and what you approve is on you. What you talk about (and anything she runs or reads for a ticket) goes to your AI provider. Early builds may still have bugs. Stick to default rules when you can, read Confirm requests, and use a separate OpenRouter key with a spend limit for beta.

Three ready-made safety setups. Relaxed and Normal allow a lot of ordinary read and write work. Riskier stuff asks first. The worst commands are blocked. They mainly differ on folders (Relaxed fully blocks places like keys and passwords; Normal and Strict also guard common system folders: writes stay blocked there, but read-only commands can still look around), how often low-risk “ask first” items run on their own, and Smart approvals (on by default only on Relaxed). Strict asks almost every time and blocks dangerous commands. You can edit the rules in Settings and reset to the defaults anytime.

It means “not a free pass.” The app will not treat that action as automatically fine. On Normal and Relaxed, a lot of low-risk confirms still run without bothering you (simple reads on both; everyday work also on Relaxed). What’s left may pop up for you. On Relaxed with Smart approvals, a second AI check may allow something that looks safe. Strict keeps asking. You can cancel or say no. From that ask you can also say “trust this kind of command for this chat.”

Known commands have their own rules. Anything else is treated as Confirm by default on all three profiles. Same idea as above: some low-risk ones may run on their own, the rest may ask you. Folder rules still apply.

Sensitive spots like SSH keys, cloud credentials, and browser login data are fully blocked on every profile, even listing them. Normal and Strict also treat common system folders as read-only: she can inspect with read-only tools, but not write or delete there. The app checks where she’s working and any folders named in the command. That is strong protection, not a perfect lockbox: if you approve a program that can wander the disk on its own, it might still reach a place the rules never saw.

Downloads need your OK by default on every profile. You can say no. Files land in a Lizzandra folder under Downloads. Getting a file is not the same as installing it. Installers still go through the same command rules as everything else.

You add an OpenRouter key in Settings so the app can talk to AI models and search. The key stays on your machine (locked with the system keychain when that works; a simpler store if not) and is only sent to OpenRouter. You do not need it for this website. For beta, a separate key with a spending limit is a good idea so a long session cannot burn your main account.

While Lizzandra is open, she notices crashes your OS already records and lists them under Events on your computer. Investigate starts a normal ticket here. Nothing quietly mails crash files to us. If she digs into a crash with you, that goes to your AI provider like any other ticket. “Start with computer” only means launch at login, tray icon, and stay in the background when you close the window. It is not a separate on/off for crash watching.

When a real website is part of the job (a download page, an account portal, a multi-step setup flow), she can open a normal browser on your computer and work through it. You can watch what she does and close the session from chat. Text she captures from a page can go to the AI so she can finish the ticket. Saving files still goes through the download rules. Uploads need your OK.

We have not paid for the official Windows and Apple signing certificates yet, so the OS may show SmartScreen or Gatekeeper warnings. That is about the installer, not about the in-app command rules. Download from this site’s official page when you can.

Public download is Windows for now. macOS works in development but is not on the download page until Apple signing is sorted. Linux is not offered publicly yet.

Start with computer: launch when you log in, show a tray icon, and keep running when you close the window. Quit from the tray when you want her fully off. Crash watching runs whenever the app is open, with or without that setting. Uninstall like any other program. On Windows your app data folder usually stays until you delete it yourself.

Right now Lizzandra is free and uses your own OpenRouter key. Pro ideas (stronger always-on monitoring and automated response) and Patreon support are plans for later. Nothing in the app bills you or unlocks a paid tier today.

Things we want later, not promises about today’s free build: scrubbing more sensitive bits before they reach the AI, stronger always-on monitoring, safer recovery options (like getting into safe mode when that helps), and deeper ways to work with the OS. Wait until they show up in the product before counting on them.

No. The desktop app is closed source.