Architecture explained without jargon
How a home AI works without the cloud.
A question can travel a few meters inside the home instead of traveling to a data center. This is the route that Famil·IA proposes and the pieces that make it possible.
First: local does not simply mean "an installed application"
An application can be installed on a computer and still send each question to an external API. When we talk about local home AI, the decisive element is where the model that generates the response is executed and where the conversation data is kept.
In Famil·IA, the model is run using Ollama on a home computer. The application coordinates the conversation with Node.js and uses SQLite for the local information it needs. The device from which the child asks the question communicates with that computer over the home network.
The journey of a conversation
1. The profile and mode are identified
The app knows which profile is active, its configured age, and whether the activity is Homework, Curiosity, Calm, or another space.
2. Applicable guidelines are met
Before generating a response, family rules are incorporated such as explaining with everyday examples, giving clues or asking for adult support in the event of a risk.
3. The model processes the request at home
Ollama runs the local model. The question and context do not need to be sent to an external provider to produce the answer.
4. The response returns to the device
The child receives an explanation adjusted to the mode. If a sensitive topic appears, the system can adapt the response and generate a notice on the family dashboard.
5. Data remains in the home environment
Conversation, configuration, and reviewable memory are maintained on the local installation.
Each layer has a different role
Interface
Presents the modes and allows you to chat from a home device.
Application
Manage profiles, guidelines, context, library, memory and alerts.
Local model
Generates the language on the home computer using Ollama.
Local data
SQLite maintains the necessary information within the family environment.
How to adapt an answer
Adaptation does not depend on a single instruction. The profile provides an age reference; the mode defines the purpose; family guidelines indicate how to help; and the context allows us to maintain coherence. A question about fractions needs a different strategy than a story about a dragon or a conversation about anger.
In Homework, clues and checks are prioritized. In Stories, creative decisions. In Curiosity, simple explanations that open new questions. In Calm, words to recognize an emotion and an everyday step, without presenting AI as therapy or replacing professional help.
What local storage provides
A familiar AI may be more useful if it remembers preferences or materials, but remembering everything automatically is not the goal. Famil·IA includes reviewable memory and a local library so that the home can decide what deserves to be preserved. The configuration also remains close to the system that applies it.
This separation means that the presentation website can be updated, indexed and shared without containing prototype conversations. They are two systems with different responsibilities.
Everyday offline operation
Once the environment is installed and the model is available, the generation of responses can continue even if there is no Internet access. The local network continues to connect the child's device to the home computer. “No internet” does not mean that devices stop communicating with each other; It means that the route does not depend on going out onto the public network.
The role of the family dashboard
The panel brings together configuration and support: age, modes, limits, personalized guidelines, reviewable memory and warnings. It is not a console to permanently observe the child. Its function is to allow family decisions to be translated into stable behavior and to facilitate an intervention when an issue really requires it.
Why build it like this?
Famil·IA tries to solve two needs at the same time. The first is privacy: that a child's conversation does not have to be sent to an external service. The second is accompaniment: that an answer does not depend only on the question, but also on the age, the purpose and the guidelines of the home.
Solving only one of them leaves the experience incomplete. A local model without adaptation may respond in a way that is unhelpful for a child. A familiar cloud-connected interface may offer guidelines, but the conversations leave home. Domestic architecture unites both decisions in the same journey.
Brief glossary
Language model
The component that generates text from the question and the context it receives.
Ollama
The tool that allows you to run the language model locally on your home computer.
Node.js
The environment with which the application coordinates profiles, modes, rules, and requests.
SQLite
The local database that allows you to preserve application information without needing an external server.
Published by the Famil·IA project. Last update: September 13, 2026.