Installation
OpenShrimp runs on your own computer — the one your projects are on, and the one that needs to stay awake for the bot to answer. Pick your platform below.
On macOS and Windows there’s an app with a setup wizard and no terminal involved. On every platform there’s also a single self-contained binary: no Python, no package manager, nothing to install alongside it.
Download the disk image, drag OpenShrimp to Applications, and open it.
-
Open the downloaded file and drag OpenShrimp into your Applications folder.
-
Launch it from Applications. The setup wizard opens on first run.
OpenShrimp lives in the menu bar (the shrimp icon) rather than the Dock. From
there you can start and stop the bot, open your config, and read the logs. The
last step of setup offers to keep it running after you sign in, and the menu’s
Start at Login changes that later — either way it registers a proper login
item, so it shows up under Login Items in System Settings. The bot runs as
its own process, so /restart from Telegram works and the menu keeps reporting
what it’s actually doing. It updates itself in place when a new release is out.
Command line
Section titled “Command line”If you’d rather run the bot from a terminal, download the binary instead of the app. Apple Silicon:
curl -fsSL https://github.com/yjwong/open-shrimp/releases/latest/download/openshrimp-macos-aarch64 -o openshrimpchmod +x openshrimpIntel:
curl -fsSL https://github.com/yjwong/open-shrimp/releases/latest/download/openshrimp-macos-x86_64 -o openshrimpchmod +x openshrimpWindows
Section titled “Windows”Download the installer and run it. It installs just for you, into
%LOCALAPPDATA%\Programs\OpenShrimp, so it never asks for an administrator
password. The bot, the .NET runtime and the Windows App SDK are all bundled —
there’s nothing to install first.
-
Run the downloaded installer. See the note below about the warning Windows shows.
-
Launch OpenShrimp from the Start menu. The setup wizard opens on first run.
OpenShrimp lives in the notification area with no taskbar window. Start and stop the bot, open your config, and read the logs from the tray menu. The last step of setup offers to keep it running after you sign in, and there’s a Start at Login toggle to change that later, backed by a logon scheduled task.
The Windows build also includes the HCS sandbox, which runs a project inside a Linux guest on the Host Compute Service. It’s optional — the bot works without it — and the guest image downloads on first boot if you turn it on.
Command line
Section titled “Command line”curl.exe -fsSL https://github.com/yjwong/open-shrimp/releases/latest/download/openshrimp-windows-x86_64.exe -o openshrimp.exeThere’s no desktop app on Linux — download the binary and run it. x86_64:
curl -fsSL https://github.com/yjwong/open-shrimp/releases/latest/download/openshrimp-linux-x86_64 -o openshrimpchmod +x openshrimpARM64:
curl -fsSL https://github.com/yjwong/open-shrimp/releases/latest/download/openshrimp-linux-aarch64 -o openshrimpchmod +x openshrimpBoth binaries need glibc 2.39 or newer — Ubuntu 24.04+, Debian 13+, Fedora 40+. On an older distribution, build from source instead.
To keep the bot running in the background and start it with the machine, see running under systemd.
Every binary, installer and source archive is also on the GitHub Releases page. The newest release is the one at the top.
Create your Telegram bot first
Section titled “Create your Telegram bot first”The setup wizard asks for a bot token straight away, so get one before you run it. It takes about two minutes in Telegram itself:
- Message @BotFather, send
/newbot, and pick a name. It replies with your bot token.
That’s the only thing you need in advance. The wizard works out who you are by having you message the bot.
Telegram Setup walks through it, and covers the two BotFather settings worth turning on while you’re there.
Run the setup wizard
Section titled “Run the setup wizard”Launch the app, or run the binary from a terminal:
./openshrimpThe wizard asks for three things:
-
Your Telegram bot token, from @BotFather. The wizard checks it with Telegram before it goes on.
-
Proof that you’re you. Search Telegram for the bot you just created and press START; it replies with a six-digit setup code. Type that into the wizard, and confirm the name it shows you. That’s what gets written into
allowed_users— nobody else’s message can take your place, and a message the bot received before you started the wizard never counts. -
Your first project — a folder on this computer, a description, and a model.
If Telegram Desktop is on the same machine, the wizard also shows a link that does step 2 in one click. And if you’d rather type an ID — you’re setting the bot up for a phone you’re not holding, say — there’s a “paste a user ID instead” escape hatch.
It writes the config to ~/.config/openshrimp/config.yaml on Linux, or
~/Library/Application Support/openshrimp/config.yaml on macOS. You can also
write it by hand — see Configuration. On
later runs it skips the wizard and starts immediately.
Sign in to Claude
Section titled “Sign in to Claude”Open your bot in Telegram and send /start. You should get a welcome message
confirming you’re connected, along with the project you’re currently in.
Now sign in to Claude. Send /login in a private chat with the bot: this opens a
Mini App that runs the same sign-in flow as the Claude Code CLI, and you paste
the resulting token back to finish. Use /login again whenever you need to
re-authenticate.
If you’d rather use an API key, set ANTHROPIC_API_KEY in the environment before
starting the bot and you can skip /login entirely. It’s read from the
environment, never from the config file.
There’s nothing else to install — the Claude Code CLI ships inside OpenShrimp. If you’d rather drive sessions with a different model, the OpenCode backend reaches OpenAI, Anthropic and Google models; it’s the one case where you install a second binary and sign in to it separately.
Next steps
Section titled “Next steps”Your bot is running and signed in. Head to First Conversation to try it out, or Configuration to add more projects.
Building from source — for older Linux distributions, or to work on OpenShrimp itself — is covered in Building from Source.