Install Blueprint as a system service

What it does

By default Blueprint serves models from the desktop GUI process — close the window, the served model stops. The Service card installs blueprint-svc as a system-managed background process:

• Windows: registered with the Service Control Manager (SCM) and started automatically at boot.
• Linux: a systemd unit under /etc/systemd/system/ set to start at boot.

Once installed, the supervisor (blueprint-svc) runs whether or not the desktop GUI is open. The GUI becomes a control panel that talks to the same supervisor; closing the window doesn't tear down the served model.

Why you'd want this

• Workstation that doubles as an LLM server: you want the model available to other apps on the machine (or to other machines on the LAN, when bound to a non-localhost address) regardless of whether you're logged in.
• Auto-restart on crash: the supervisor restarts llama-server with exponential backoff when it dies. The desktop-only path loses the model and waits for you to click Start again.
• Survives reboots: laptop closes, machine sleeps, Windows decides to install updates — the service comes back up automatically.

If you only run models interactively during your dev session, you don't need this. The Deploy tab's serve loop is fine.

How to install

Windows

1. Open Dashboard → Maintain.
2. In the Service card, click Install service.
3. Windows prompts for UAC elevation — click Yes. The installer copies blueprint-svc.exe to %ProgramData%\Blueprint\ and registers it with the SCM.
4. The card flips to "Installed · Stopped". Click Start managed server to bring it up; the SCM state field flips to "Running" once the supervisor binds the control plane.

Linux

1. Open Dashboard → Maintain.
2. In the Service card, click Install service.
3. You'll be prompted for sudo (the installer needs root to write /etc/systemd/system/blueprint-svc.service and copy the binary to /usr/local/bin/).
4. Once installed, the service is enabled and started by the same click. systemctl status blueprint-svc from your terminal should show it active.

Apply a serve config

Installing the service doesn't start serving a model — it starts the supervisor, which waits for a config. In the Service card:

1. Pick a model + quant from the dropdown (uses the same catalog the Plan tab does).
2. Tune the context size and GPU layers if you need to — the defaults work for most setups.
3. Click Apply config. The supervisor picks up the new config within ~5 seconds and spawns llama-server against it.

The card shows the current phase: idle → starting → running → on crash, crashed with LastError and a restart countdown.

Uninstall

Same card has an Uninstall service button. Stops the supervisor, removes the SCM entry (Windows) or systemd unit (Linux), and deletes the installed binary. Your models and configs stay on disk; only the service registration is removed.

Common pitfalls

• Port conflict at 127.0.0.1:17832: another instance of blueprint-svc is already bound (left over from a debug run, or an older install). Check with netstat -an | grep 17832 and kill the squatter.
• "Service installed but won't start": check the log at %ProgramData%\Blueprint\logs\ (Windows) or journalctl -u blueprint-svc (Linux). Most often: the model GGUF has been moved or deleted and the supervisor can't find it.
• GUI shows "service unavailable" after install: the GUI talks to the service via the bearer token at ~/.blueprint/svc-token. If you uninstalled and reinstalled, the token rotated and any remote hosts caching the old token in their OS keychain need to reconnect (which triggers a refetch).