Platforms

Platforms and Deployment Modes

FoodAssistant can run as a server stack, as a self-contained Raspberry Pi appliance, or as a thin Pi control surface for a stack that lives elsewhere. This page describes the three deployment modes, how to host the server stack, the AI providers you can plug in, and the Home Assistant integration.

For the physical hardware and peripherals, see Hardware.

Deployment modes

The mode is chosen in the setup wizard (or carried in via config.env on a flashed device) and stored in the app's settings.json. The three modes are defined in service/app/config.py:

• server. FoodAssistant runs on a general server (NAS, mini PC, VM, and so on). It connects to a separately running Grocy, and optionally Mealie. This is the only non-Pi mode.
• pi_hosted. Everything runs on one Raspberry Pi: FoodAssistant plus Grocy, and Mealie by default, with or without an attached display. The kiosk auto-enables when a display is present, and a display can be added later. The first-boot provisioner also installs a host bridge and a port 80 to 9284 redirect so the device is reachable at http://<hostname>.local/.
• pi_remote. A thin satellite. The device drives a kiosk and/or Stream Deck pointed at an existing FoodAssistant server on the LAN. There is no local Docker, Grocy, or Mealie; the satellite runs the FoodAssistant UI on port 80 (via a Python venv, or optionally Docker) and pulls its backend config from the main server. This is what makes a low-spec board like a Pi Zero useful.

For exactly which settings are editable, inherited from the server, or device-local in each mode, see the Settings visibility matrix.

Updates across the fleet

A single "Install updates automatically" setting (on by default) drives updates for the whole deployment:

• A server install applies updates through the bundled Watchtower container, which checks for a new FoodAssistant image and recreates the service container when one is published.
• A pi_hosted appliance applies updates through the host-bridge over-the-air helper, and the in-app update control on its Settings page can check and apply on demand.
• A pi_remote satellite inherits the flag from its main server (it is one of the SATELLITE_PULL_FIELDS), so a server and its satellites converge on the same version rather than drifting apart.

Hosting the server stack

The server and pi_hosted modes run the stack with Docker and Docker Compose v2. Optional backends are gated behind compose profiles, so you only run what you need.

Profiles:

• (default, no profile or with-grocy): the FoodAssistant service plus Grocy.
• with-mealie: adds Mealie for recipes, meal plan, and shopping list.
• with-ollama: adds Ollama for fully local AI.

Example enabling everything:

docker compose --profile with-grocy --profile with-mealie --profile with-ollama up -d

Pinned backend versions and ports

The bundled backends are pinned to specific image tags (not :latest) so an unattended pull cannot move you onto a breaking release.

Service	Image	Tag	Port
FoodAssistant	ghcr.io/syracuse3dprinting/foodassistant	${FOODASSISTANT_TAG} (default latest)	9284
Grocy	lscr.io/linuxserver/grocy	4.6.0	9383
Mealie	ghcr.io/mealie-recipes/mealie	v3.19.2	9285
Ollama	ollama/ollama	0.30.8	11434

To move a backend to a newer version, back up first, then bump the tag in your compose file and recreate just that service.

Reverse proxy and URL caveat

You can put FoodAssistant behind a reverse proxy (for example Pangolin) to get a public URL. One important caveat: headless clients must use the LAN URL, not the public proxy URL. A request without a browser session (for example a Home Assistant REST sensor) hitting the public URL gets an HTML redirect rather than the JSON it expects. So Home Assistant REST sensors point at the LAN address such as http://192.168.1.170:9284, while human-facing Lovelace buttons can use the public URL.

AI providers

AI features are optional. The vision provider is selected in the setup wizard. The supported providers (see vision_provider in service/app/config.py):

• gemini (Google). Cloud, API key. The default provider.
• openai. Cloud, API key.
• anthropic. Cloud, API key.
• ollama. Fully local inference, no API key. Reads from an Ollama instance (bundled via the with-ollama profile on port 11434). Local vision models are heavy, so this is best on x86-64 with plenty of RAM rather than a small SBC.

If you do not want local inference, configure any one of the cloud providers and any supported board is fine.

Home Assistant integration

There are two ways to use FoodAssistant with Home Assistant:

• Add-on (HA OS / Supervised). FoodAssistant installs as an add-on and lives in the HA sidebar with no separate login; HA authenticates the UI through Ingress.
• Standalone. A standalone FoodAssistant instance exposes REST endpoints that HA can consume. The homeassistant/ directory ships REST sensors (expiring summary, inventory dashboard, pending scan count, and the Mealie shopping and meal-plan summaries), a rest_command for posting scans, automations, and a Lovelace dashboard.

A common HA pattern is the headless barcode scanner: a USB or Bluetooth HID scanner is captured with the keyboard_remote integration, an automation buffers the typed digits, and a rest_command posts the completed barcode to /pending/scan on the FoodAssistant LAN URL. keyboard_remote only works on Home Assistant OS / Supervised. See homeassistant/barcode-scanner.md for the full walkthrough, and remember the LAN-URL caveat above for all REST sensors and commands.