Start here

The apps

You do not need a terminal to use llamay. Each app below puts a chat window in front of the same engine, runs the model on your own machine, and keeps your conversations on it. Pick the one for the computer in front of you.

Every file is on the download page. An Intel Mac runs the command line and Studio, not the Mac app, which is built for Apple silicon only.

Your first conversation

1. Install and open the app. The engine is inside it; there is nothing else to install.
2. Get a model. The first run offers a small one to download. On the Mac it is Qwen 2.5 0.5B, about 400 MB, which runs on any Apple silicon Mac.
3. Ask it something. The answer is generated on your machine. Turn the network off and it still answers.

macOS

Open the .dmg and drag llamay to Applications. It is signed and notarised by Apple, so it opens without a warning. It needs macOS 13 or later on Apple silicon.

llamay lives in the menu bar. From there: open the window, Quick Entry for a question from anywhere, start or stop the server, Install Command Line…, open at login, keep the server running at login, check for updates, and show the log.

• The command line. Install Command Line… links llamay and llamay-server into /usr/local/bin, after asking for an administrator password. The app and the command line then share one model store, ~/.llamay/models.
• Where things are kept. Conversations are in ~/Library/Application Support/llamay/. The sign-in token is in the Keychain. The log is ~/Library/Logs/llamay.log.
• Updates. The app checks for a new release once a day, and asks before installing unless you turn on automatic installs. A download is checked against the release's SHA256SUMS, its code signature, and the signing team before it replaces the app.

Windows

Run the setup program. It installs for you alone by default, into %LOCALAPPDATA%\Programs\llamay, so it asks for no administrator rights. It offers to add llamay to your PATH and to start the server when you sign in.

The Start menu then has two windows onto the same engine:

• llamay shows Studio in a window, with a tray icon. Alt+Space asks a question from anywhere. It needs Microsoft's WebView2 runtime, which Windows 11 already has; if it is missing, the window says where to get it.
• llamay (native window) is a native Windows app with settings panes for appearance, account, updates, the engine's log, a seat licence, a Hugging Face token, providers, memory, files and sync. Its settings live in %LOCALAPPDATA%\llamay\, and secrets are protected with Windows DPAPI.

Unattended installs, /ALLUSERS and the logon task are covered on the install page.

Linux

The desktop window is a separate package, llamay-desktop, that depends on the engine package llamay and on WebKitGTK. Install both from the same release:

sudo apt install ./llamay_<version>_amd64.deb ./llamay_<version>_linux_amd64_desktop.deb
# Fedora: sudo dnf install ./llamay-<version>-1.x86_64.rpm ./llamay_<version>_linux_amd64_desktop.rpm

The applications menu then has llamay. It runs llamay app, which opens the window when there is a display and a browser tab when there is not. The engine package also installs a system service; see the service the packages install.

Studio, in any browser

Every llamay serve serves Studio at /studio. It is a chat window, a model picker that searches Hugging Face, memory, and connectors. llamay app starts a server if none is running and opens it:

llamay app                 # start the server if needed, open Studio
llamay app -no-open        # start it, print the address, open nothing

• It answers this machine only. Studio accepts requests addressed to localhost, a loopback address or a .local name. A request for any other host name gets 421 Misdirected Request, so a server on a network does not hand Studio to the network.
• Your history is in the browser. Conversations are kept in that browser's storage for that address. A server on a different port shows an empty history.
• A folder to work in. llamay serve -studio ~/project gives Studio a file tree over that folder. -studio-exec also lets it run commands there. Neither is on by default.

Dictation

Studio has a microphone button. Start the server with a Whisper model and dictation is transcribed by llamay on this machine:

llamay serve -whisper ggml-base.en.bin

Android

The Android build is the command line for arm64 devices, with no installer and no app. Unpack it in a terminal on the device and run it like any other llamay. It runs on the CPU only: it has no GPU backend. It is built and shipped with every release, but these docs have not tested it on a device.

tar xzf llamay_<version>_android_arm64.tar.gz
./llamay pull qwen2.5:0.5b
./llamay run -m qwen2.5:0.5b -chat -p "hello"

What the apps send over the network

A conversation with a model on your machine sends nothing. These are the exceptions, and each is something you can see and turn off:

• Update checks, once a day, to GitHub's release API for the public llamay-releases repository. LLAMAY_NO_UPDATE_CHECK=1 stops the check in the command line, the server and the Windows and Linux windows. The Mac app has its own setting, Check for Updates Automatically, in the menu bar.
• Downloading a model goes to wherever the model is: Ollama's registry, Hugging Face, or the URL you gave.
• Signed in, the apps talk to llamay's hosted service for sync, hosted models and plans. Signed out, they do not.
• A hosted model sends that conversation to llamay's service. A provider you add sends it to that provider. Local models send nothing.

Where to go next