Get Shell:B

A private AI workspace that runs on your own hardware. Install the app, add the helpers you want, or run your own.

Open Shell:B

Install the app

Shell:B installs straight from the browser and then runs in its own window, with notifications, file sharing and shellb:// links.

Chrome or Edge

Open ai.shellb.app, sign in, then choose the install icon at the right of the address bar (or Settings → General → Install).

Safari on a Mac

Open ai.shellb.app, then File → Add to Dock.

iPhone and iPad

Open it in Safari, tap Share, then Add to Home Screen.

Android

Open it in Chrome, tap ⋮, then Install app.

Desktop helpers

Small optional programs for your own computers. They talk to Shell:B over your private network (Tailscale), not through this site.

Screen

Lets Shell:B look at this computer's screen and use a browser window on it, only when an agent asks and only while it runs. The easiest copy comes from Settings → Tools & Connectors → Screen → Pair a computer, with the address and token already filled in.

pip install mss pillow playwright
python shellb_screen.py --url https://<device>.<tailnet>.ts.net:8443 --token sbs_…
shellb_screen.py

Fleet node

Lets Shell:B run its models on this machine's Ollama too, so the load spreads. Get a token from Models → Fleet → Add a machine. Python 3.9+, no other packages.

python3 shellb_node.py --token <token> --install-service
shellb_node.py

shellb:// links

Makes shellb:// links open in the installed app. Install the app first.

Linux

mkdir -p shellb-link && cd shellb-link
for f in shellb-link shellb-link.desktop install-linux.sh; do curl -fsSLO https://get.shellb.app/clients/shellb-link/$f; done
sh install-linux.sh

macOS

curl -fsSLO https://get.shellb.app/clients/shellb-link/install-macos.sh && sh install-macos.sh

Windows

Download shellb-link.reg and double-click it (per-user, no admin).

API and MCP

Create a key in Settings → Tools & Connectors → API, then choose which agents it may use.

REST and OpenAI-compatible
https://api.shellb.app/api/v1
MCP (streamable HTTP)
https://api.shellb.app/mcp
Auth
Authorization: Bearer sbk_…
curl https://api.shellb.app/api/v1/chat/completions \
  -H "Authorization: Bearer sbk_…" -H "Content-Type: application/json" \
  -d '{"model": "shellb", "messages": [{"role": "user", "content": "Hello"}]}'

Full API reference

Run your own

Shell:B is open source: FastAPI and React, with local models through Ollama. It's built for an NVIDIA DGX Spark but runs on any Linux machine with a capable GPU.

Shell:B on GitHub