Home / Docs / CLI reference

CLI reference

TurboLLM runs from a single command. Start the daemon, bind it to your LAN, expose it over a tunnel, stop it from any terminal, or launch a coding CLI wired straight to your local models — every flag is listed below.

Start the daemon

With no arguments, TurboLLM starts the daemon on port 6996 and opens the web UI in your browser.

$turbollm

Everything else is a flag on that same command. Point at a different port, skip the browser, or bind to a specific address:

turbollm --port 9000            # listen on a specific port
turbollm --no-open              # start without opening a browser
turbollm --addr 0.0.0.0:6996    # bind all interfaces (LAN sharing)

Flags

FlagWhat it does
--port <n>Port to listen on. Default 6996.
--addr <host:port>Bind to a specific host and port, e.g. 0.0.0.0:6996 to expose on your LAN.
--no-openStart the daemon without opening a browser window.
--tunnelExpose the daemon on the internet via a cloudflared quick tunnel. Prints a public URL and a required access token — see Cloud Launch below.
--config <file>Use a specific config file instead of the default.
--stopStop a running daemon. Reads ~/.turbollm/daemon.pid, so it works from any terminal.
--help, -hShow usage and exit.

Stop a running daemon

You don't need the terminal that started TurboLLM to stop it. --stop reads the pid file at ~/.turbollm/daemon.pid and shuts the daemon down from anywhere.

turbollm --stop

LAN sharing

To use TurboLLM from other devices on the same network — a laptop, a phone, a second workstation — bind it to all interfaces instead of localhost.

  1. Bind to all interfaces

    Start with --addr 0.0.0.0:6996 so the daemon accepts connections beyond localhost.

  2. Open it from another device

    On the same network, browse to http://<your-ip>:6996.

  3. Require an API key

    Turn on Require API key in Settings → Network whenever you expose the daemon, so only authorized requests are served.

turbollm --addr 0.0.0.0:6996

Cloud Launch

Running TurboLLM on a rented cloud GPU box? --tunnel opens a cloudflared quick tunnel and prints a public URL you can reach from anywhere, plus a required access token. Only requests carrying that token are served, so the endpoint isn't open to the world.

$turbollm --tunnel --no-open
Keep the token

The tunnel serves only requests that carry the access token printed at startup. Copy it from the output — without it, the public URL returns nothing.

Launch a coding CLI

The launch subcommand wires a coding CLI to TurboLLM and starts it in one step. If no model is running, launching Claude Code auto-loads one for you.

$turbollm launch claude

Other coding CLIs work the same way — launch points them at your local TurboLLM endpoint, then starts them:

turbollm launch claude      # start Claude Code (auto-loads a model if none is running)
turbollm launch opencode    # wire opencode (or kilo / openclaw / hermes), then launch

See Integrations for the full list of launch targets and setup notes, and Configuration for the environment variables TurboLLM reads.

Previous← Anthropic-compatible