BOAT
Download

Using BOAT

Configuration

Ports, network binding, upload limits, the default device, autostart and release channels.

boat setup is the single source of configuration. It writes config.toml in BOAT's config directory, validates it, checks that the ports are free and restarts BOAT if it was running.

Guided setup

boat setup

The interactive terminal UI walks through each setting. Setup never installs models; install those separately with boat model install.

Unattended setup

Every setting has a flag. Pass --non-interactive to apply them without the UI, for example in provisioning scripts:

boat setup --non-interactive \
  --bind 127.0.0.1 --port 7234 --backend-port 7236 \
  --upload-limit 1g --default-device auto \
  --autostart=false
Flag Default Description
--bind 127.0.0.1 Address the web app listens on
--port 7234 Port for the web app
--backend-port 7236 Internal port for the backend API
--upload-limit 1g Maximum upload size, such as 500m or 2g
--default-device auto auto, cpu, mps or cuda:<n>
--autostart true on Linux Start BOAT at login through user systemd
--channel stable Release channel for boat update: stable or beta
--accept-insecure-lan off Required with a non-loopback --bind

Network access

BOAT listens only on 127.0.0.1 by default, so nothing else on your network can reach it. To let teammates on your LAN connect, bind to another address and explicitly accept the risk:

boat setup --non-interactive --bind 0.0.0.0 --accept-insecure-lan

Plain HTTP is not for the internet. Serving BOAT directly on the internet is not supported. Keep the default loopback binding, or put BOAT behind a reverse proxy that terminates TLS.

Default device

The Default Device decides where models run unless a model has its own setting.

Value Meaning
auto MPS on Apple silicon. On Linux, CUDA 12.6, CUDA 12.8 for Blackwell GPUs, or CPU when no NVIDIA GPU is present
cpu Always use the CPU
mps Apple silicon GPU
cuda:<n> A specific NVIDIA GPU by logical index, e.g. cuda:0

If an NVIDIA GPU is present but its driver or CUDA probe fails, BOAT reports the failure instead of silently falling back to the CPU. boat doctor --json shows the active runtime profile, eligible GPU UUIDs and accelerator status.

Changing the Default Device re-applies it to installed models. Individual models can override it with boat model setup; see Models.

Autostart

On Linux, BOAT uses a user systemd service, enabled by default for release installs when systemd is available. macOS uses a fallback supervisor and does not start at login by default.

boat service enable    # start at login and after reboot
boat service disable

Inspect and validate

boat config show       # print config.toml
boat config validate   # check the config and device settings

Local administration, such as creating users or resetting passwords, goes through the CLI and is protected by your user account's filesystem permissions.

↑↓ navigate↵ open