Manual setup

Get the source and install dependencies

Clone the repository into a location accessible to the account that will run the service. For example:

git clone https://github.com/ndom91/llama-dash.git /opt/llama-dash
cd /opt/llama-dash
pnpm install --frozen-lockfile

The destination must be writable by your installation account. To pin a build, check out a specific Git commit before installing dependencies.

Configure the environment

cp .env.example .env

Edit .env before starting:

• Set INFERENCE_BASE_URL to your llama-swap address, for example http://127.0.0.1:8080. The example file's http://llama-swap:8080 address is for Docker Compose.
• Set BETTER_AUTH_SECRET to the output of openssl rand -base64 33. Keep .env readable only by the service account and administrators.
• Set BETTER_AUTH_URL to the browser-visible HTTPS origin when deploying behind a reverse proxy or using passkeys remotely. It can be left empty for a local install.
• To use the dashboard's Config page, set INFERENCE_CONFIG_FILE to the absolute path of the config.yaml your llama-swap instance loads. This is llama-swap's config file, not a separate llama-dash config. Otherwise, clear the example's Docker-specific /config/config.yaml value.
• Set CREDENTIAL_ENCRYPTION_KEY to a secret of at least 32 characters before storing upstream credentials.

The service account needs write access to the database directory and, if using the config editor, the llama-swap config directory. The default database path is data/dash.db, relative to the working directory. Migrations run automatically at startup. See Configuration for the full list of variables.

Build and start

pnpm build
pnpm start

The production server listens on port 3000 on all interfaces by default. Stop the foreground process with Ctrl+C before starting it under systemd. For remote access, use HTTPS via a reverse proxy and restrict access to the service as appropriate.

Run as a systemd service

Create /etc/systemd/system/llama-dash.service. Replace llama with an existing non-root service account, /opt/llama-dash with your checkout path, and /usr/local/bin/pnpm with the absolute path from command -v pnpm. Make sure Node is also on the service's PATH; a version-manager installation may require an explicit Environment=PATH=... in the unit.

[Unit]
Description=llama-dash inference gateway
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
User=llama
WorkingDirectory=/opt/llama-dash
Environment=NODE_ENV=production
Environment=HOST=127.0.0.1
Environment=PORT=3000
ExecStart=/usr/local/bin/pnpm start
Restart=on-failure
RestartSec=5

[Install]
WantedBy=multi-user.target

The example binds to loopback for use with a local reverse proxy. Change HOST only if the server should listen on another interface. HOST and PORT can be set in the service environment; the working directory lets the app find .env, the database, and migrations.

Ensure the service account can read the checkout and .env and write the configured database directory. Then enable the service and inspect its status:

sudo systemctl daemon-reload
sudo systemctl enable --now llama-dash
sudo systemctl status llama-dash
sudo journalctl -u llama-dash -f

Open http://localhost:3000 locally or your configured HTTPS origin, create the first dashboard user, and connect a client. To update later, see Updating.