A private AI assistant for the whole family
This page may contain Amazon affiliate links. A purchase after a click can earn the site a commission, at no extra cost to you. The advice stays independent. How affiliation works.
A private family assistant is a chat that runs in the house. Ollama runs a language model on your computer. Open WebUI is the chat page in the browser, readable for adults and teenagers. Conversations stay on that machine as long as you leave the server off the public internet. This guide is a 30-minute setup plus one concrete job: a parent summarizing school email.
What you keep, and where it stays limited
| You keep | It still has limits |
|---|---|
| Control of what gets typed | The model can hallucinate (make things up) |
| Room for the whole household to chat | Flagship cloud models are still sharper |
| A long-term cost of hardware plus kilowatt-hours | Someone has to install updates |
Model updates on your schedule (ollama pull when you choose) |
A cloud plan refreshes the model on its own |
| Prompts that stay home | Risk if port 8080 is open, or if guest Wi-Fi can reach the server |
For the money over several years, read the 3-year cost of Ollama and ChatGPT.
Hardware that is enough
- A quiet mini PC with 32 GB of RAM (noise and power, mini PC comparison).
- A 512 GB to 1 TB SSD (model storage).
- An 8B Q4 model runs on the processor. A GeForce RTX helps when several people chat at once.
If you still need a dedicated machine, the 32 GB profile above matches this guide: an 8B Q4 chat for the whole family. Partner search on Amazon.com, with the model left to you: compare results with the hardware selector and the comparison guides. 32 GB mini PCs on Amazon.com. Check the current price on the store (October 2026). How affiliation works on LocAIhome.
A 30-minute setup
1. Ollama
Follow install Ollama on Windows or the Windows, Linux, and Mac install. Then test:
curl http://127.0.0.1:11434/api/tags
2. A model the family can talk to
ollama pull mistral:7b-instruct-q4_0
Use the exact tag from the Ollama library that day. What matters is an instruct model at Q4 (choosing a first model). Swap the tag if another small instruct model writes more naturally for your household.
3. Open WebUI
Install it with the beginner Open WebUI guide. Open http://localhost:8080 and create the administrator account. The first account is the admin.
4. Family accounts
| Role | Habits that help |
|---|---|
| Admin (a parent) | A long password, and a backup of the settings |
| Teenager | A separate account, and a reminder that the model is a tool |
| Younger child | A parent nearby, a time limit, and no sensitive details |
5. Try it on a school email
- Paste a school email into Open WebUI. For a test, leave out full names.
- Prompt: “Summarize this in 5 bullets and list the important dates.”
- Check every date yourself before you act on the summary.
- Voice-message variant (a voicemail or an email with an audio attachment). Reuse this same chain later in the guide:
- Export the audio as .ogg or .mp3 on the family computer.
- Whisper (offline speech-to-text) turns it into words. Use first steps with Whisper or the longer offline Whisper guide.
- Prompt Ollama in Open WebUI: paste the transcript and ask for “5 bullets, and the dates written out.” The reply stays on the machine.
If a mini PC with no GPU feels slow, try a 3B instruct model for the children and keep the 8B for the parents (choosing a first model). The hardware selector helps you decide whether an RTX is worth it for several chats at once.
A parent checklist: privacy, children, and the network
One list for adults: children’s data, open ports, and guest Wi-Fi. Read it next to privacy when the AI stays home and the CNIL page on children’s data (France’s data-protection authority; the page is in French).
| Topic | What to do |
|---|---|
| Children’s data | Skip report cards that include full identifiers. Leave health details and contact info out unless you truly need them |
| Internet exposure | Leave ports 8080 (Open WebUI) and 11434 (Ollama) closed on the router |
| Tablet access | Home network only: the PC’s local address, and a public web address stays unpublished |
| Guest Wi-Fi | A separate network name for friends and visitors, with no route to the AI server’s address |
| Backups | Encrypt conversation exports if you archive them |
| Updates | Plan OS, Ollama, and Open WebUI updates. Run ollama pull after you read the release notes |
Software that fits next to the chat
- Whisper turns an audio file (a school voicemail, a voice message) into text before Ollama summarizes it.
- n8n can watch a folder, run Whisper, then send a prompt to Ollama, so you stop clicking through each file. That is an intermediate setup.
- Home Assistant if the same house also runs the lights and sensors.
Limits to explain to children
The model can invent facts, offer risky “advice,” or sound sure of itself. Medical and legal questions wait for an adult. It is a study aid. A teacher and a doctor still do their own jobs.
Cost and electricity
A machine left on all day: estimate the kilowatt-hours with the electricity budget. A ChatGPT subscription is optional when the local chat is enough for the household.
Scenarios at a glance
| Who | Typical use | Reminder |
|---|---|---|
| Parents | Paste school email into Open WebUI and ask for bullets | Check the dates by hand. Voice messages use step 5 |
| Teenagers | Understand a lesson. Handing in the model’s wording as the homework is the habit to stop | Code lives on a separate computer: Continue and Ollama |
| Guest Wi-Fi | The AI server stays unreachable | A separate guest network. See the parent checklist above |
The walkthrough is in the next section. The house-rules appendix at the end covers the written charter and Docker backups.
Family scenarios in practice
Parents: email and schedules
An 8B instruct Q4 model summarizes text you paste into Open WebUI. A parent still checks the dates before anyone acts on them. For a voice message (a messaging app, or an audio attachment), use the same chain from step 5: export the audio, run Whisper, prompt Ollama. A second tutorial is unnecessary.
Teenagers: homework and code
Set the rule in plain language: the model helps them understand, and the assignment is still theirs. For code, set up Continue (an editor extension pointed at Ollama) on a separate computer. See the local coding guide. That keeps teenage Open WebUI accounts separate from the coding setup.
Updates on a calm schedule
Once a month, pick a window: ollama pull on the family model, back up the Docker volume if you use one, and run a quick test such as “say hello.” If the chat gets slow after an update, check VRAM and the size of the model that is loaded.
Local chat and a cloud subscription, side by side
| Cloud chat (a Plus-style plan) | Ollama and Open WebUI at home | |
|---|---|---|
| What you type | On the company’s servers | On your disk, or in the Docker volume |
| How much you can use | The company’s fair-use rules and quotas | Whatever the RAM and GPU can carry |
| Cost over 36 months | The subscriptions added up | The computer plus kilowatt-hours |
| Model updates | Automatic | You choose the pull |
The dated ranges and named scenarios are in Ollama vs ChatGPT over 3 years. This page stays on setup and on who is allowed to use the chat.
Useful next steps
- A first local AI in about 15 minutes, before you open Open WebUI to the whole family.
- Docker for Ollama and Open WebUI, if you prefer containers.
- VRAM · 32 vs 64 GB of RAM · GPU, iGPU, eGPU, and OCuLink.
- Images without a cloud, a different job from text chat.
- Continue, a local coding assistant, for teenagers who write code.
- Choosing hardware · the selector · a Mac mini M4 if you are already on Apple.
Appendix: house rules for the family computer
- A short charter posted by the PC: health and legal questions wait for an adult, and the model can be wrong.
- Passwords: one account per teenager, changed every quarter. The parent admin keeps the backup of Open WebUI exports.
- Logs you can actually read: an occasional JSON export of conversations, so you can spot risky use.
- Offline chat: it works after the models are downloaded. Keep OS and Ollama updates on the calendar anyway.
- Two admin accounts: two parents as admins, so a lost password does not lock the household out.
- Docker backup: if you use Docker with Open WebUI, back up the
openwebui_datavolume. - A lighter model for children: a 3B instruct model answers faster on a mini PC. See choosing a first model.
- Whisper and audio email: reuse step 5, or the Whisper and n8n stack above. A third procedure is unnecessary.
- Images: separate rules if teenagers move on to ComfyUI or images without a cloud.
- Storage and the power bill: an SSD for models and the electricity budget if the machine stays on all day.
Common questions
Does the chat need the internet the whole time?
Once the models are downloaded, chatting works offline. You still want the internet for Ollama updates, Open WebUI updates, and operating-system patches.
How do several family accounts work?
Open WebUI supports more than one user. A parent administrator creates the accounts, each person gets a unique password, and everyone hears the same reminder: a language model can be wrong.
Which model should a family start with?
An instruct model around 8B, quantized to Q4, that follows everyday language well. See choosing a first model. 32 GB of RAM is the comfortable starting point.
Are children’s messages protected?
A local chat skips the usual upload to a commercial cloud. Stay careful about what children type, and keep the server off the public internet.
Sources
Narrow it down with the hardware selector · French version of this section