Download#

Download the latest version (free, MIT licensed)

One file for your system, from the table below or the Assets list of the release. Nothing else to install: Node.js and the character creator are inside. You still need a brain (below); without one she runs a small built-in script and tells you how to get one.

Signing the Windows installer, and notarising the Mac app, may come later; until then Windows and macOS warn as described below.

The first start#

The app is not signed with a developer certificate (that needs a paid account), so your system warns you once. It is the same program you can read in this repository.

• Windows: SmartScreen says "Windows protected your PC". Choose More info, then Run anyway. On Windows 11 with Smart App Control switched on, an unsigned installer is blocked with no way to run it anyway: turn Smart App Control off (Windows Security → App & browser control → Smart App Control) to install. The installer is per user and needs no administrator.
• macOS: open the .dmg, drag the app into Applications, then right-click it → Open, and Open again. (Or System Settings → Privacy & Security → Open Anyway.)
• Linux: make the file executable and run it: chmod +x Anime.AI.Companion-0.20.0.AppImage and double-click it, or start it in a terminal. An AppImage uses FUSE 2 (libfuse2, or libfuse2t64 on Ubuntu 24.04 and newer). Without it, newer AppImages unpack themselves into a folder in /tmp and start from there (this prints a long file listing and leaves that folder behind); an older one fails with error loading libfuse.so.2. Installing that package avoids both, or unpack the file yourself and run it from a folder of your own: ./Anime.AI.Companion-0.20.0.AppImage --appimage-extract, then squashfs-root/AppRun. The launcher entry the AppImage makes starts the app with --no-sandbox (an AppImage cannot carry the setuid helper Chromium's sandbox wants); started by hand, add it yourself if you see an "unprivileged user namespaces" or chrome-sandbox error. The folder picker needs zenity or kdialog, and the tray icon needs a desktop that shows one. --version prints the version and --help the options.

The release also has a SHA256SUMS file; sha256sum -c SHA256SUMS --ignore-missing checks what you downloaded.

What you need besides the app#

One brain, the same ones the Brains page describes: Claude Code (claude, signed in), opencode, Cursor (cursor-agent) or Ollama. The app looks for them the way your terminal does, so a brain you can start from a terminal is found. It picks one it finds on the first start; after that the choice is yours, in the page.

Where things are#

Her settings, chat, look and the folder she works in are in your user folder, never inside the app, so installing a newer version keeps them:

• Windows: %APPDATA%\Anime AI Companion
• macOS: ~/Library/Application Support/Anime AI Companion
• Linux: ~/.config/Anime AI Companion

Closing the window hides her to the tray or menu bar; she and her server keep running until you choose Quit.

Updates#

Nothing is installed behind your back. The first time, the app asks whether to look for a newer version about once a day; if you say yes it asks GitHub for the newest version number and GitHub Pages for the list of picture packs, to see whether packs you downloaded have newer versions (they see your IP address and that the app asked, nothing else), and tells you when there is one, with a button to this page. Newer packs are only offered; the app asks before it fetches one. The app asks only once; the switch is under File → Options, and the Help menu has Check for Updates….

Speaking to her#

The microphone button needs your browser's speech recognition, which the app's window does not have. Its button offers to open the same page in your browser, where it works (and still asks first).

Prefer to run it from source?#

Getting started has the clone-and-run route, the anaico command for work mode in a folder, and every setting. Or try the demo first: nothing to install.

Improve this page on GitHub

Made with AI coding agents, mostly Claude Code. This documentation was written by Claude Code. Details