Skip to content
Ambrogio

What is Ambrogio

Talk to it like Alexa, chat with it like ChatGPT: Ambrogio is one self-hosted AI assistant for your whole homelab. Multi-agent by design: each agent you configure handles one job, such as the house, music, weather or your documents, using the services you already host. Each agent uses an LLM from a cloud provider or one that runs in your home, and conversations are stored on your server.

It can take the place of a voice assistant such as Alexa: it listens in every room for its wake word and answers out loud. It also goes further. You can type to it from a browser or a terminal and continue the same conversations, it runs on your server, and it works with the services you choose instead of a vendor’s.

Ambrogio does not do these things by itself. Each agent can be connected to a service that already runs in your home, such as Home Assistant or a file cloud, through the Model Context Protocol, and the orchestrator chooses the right agent for each request.

That is why Ambrogio is designed as the AI part of a homelab: it sits beside the services you already self-host and lets the family use them by talking. The ambrogio-homelab-example repository shows a complete installation of this kind, with Home Assistant, a weather service, a file cloud, web search and a voice satellite.

Four programs

Section titled “Four programs”

Ambrogio is split into four programs, each in its own Git repository. You always need the backend; the clients are optional, and you can run as many of them as you like.

The backend (ambrogio-backend) runs on a machine that is always on, such as a home server or a NAS with Docker. It keeps every conversation in a PostgreSQL database that runs beside it. It sends each request to an orchestrator, which picks the agent that should answer, and to that agent, which writes the answer; each agent has its own instructions and its own model. It recognises speech while you speak and speaks the answers, with models that run inside the backend and no cloud speech service. Clients reach it through two HTTP endpoints, to read conversations, and a WebSocket session, to talk.

The web chat (ambrogio-web) is a chat page for the browsers of your home, served by an optional container next to the backend. It shows the answers as Markdown, in Italian or English.

The terminal client (ambrogio-cli) is the same chat in a terminal window, for any computer that can reach the backend.

The voice satellite (ambrogio-satellite) is a program for a small device with a microphone and a speaker, such as a Raspberry Pi, one per room. It waits for its wake word, “Alexa” unless you choose another, sends what you say to the backend and plays the answer. It does not recognise speech or generate voices itself.

One family, one archive

Section titled “One family, one archive”

Everyone in the home shares one identity, and every conversation can be opened from any client: a question asked out loud in the kitchen can be continued by typing in the browser. Ambrogio has no user accounts and no login, so it belongs on your home network; Security and privacy explains what that means and what leaves your home.

Further reading

Section titled “Further reading”