AI for my 10-year-old son
This is a follow-up to
Geoffrey Huntley
but targeted more towards parents. My son recently turned ten, and after school, we have been building (and authoring blog posts together) with/about AI. The industry is changing fast, and he has eight years ahead of him before he becomes an engineer in the workforce.
He's always liked to tinker, and after speaking with some mates who have done similar for their kiddos, here are my notes on how I provisioned a dedicated child-safe instance of OpenWebUI for my son so that he can upskill himself.
what is open-webui?
OpenWebUI is a self-hosted multi-llm chat interface that can be deployed on your own infrastructure. Instead of purchasing a subscription for ChatGPT + Claude—circa $40 USD per user—you can run an OpenWebUI instance with unlimited users and pay per query.
OpenWebUI can also be customized with system prompts on a per-user basis. See below for personalization and a screenshot of a single query being fanned out to one or more LLM models.
how can you do this yourself
- Install https://github.com/open-webui/open-webui and self-host it.
- Register for https://openrouter.ai/ and obtain an API key.
- Create an account for your 10, head to "Settings -> Memory -> Personalization", and use a system prompt like the following on the account...
The system replies in the format where the AI is a Panda. The AI's name is "Bamboo" and she occasionally tells funny jokes. Bamboo is a rascally Panda that wants to become a millionare and desires to take over the world.
<child_safety>
You must ensure all content is appropriate for children aged 10 and under. This means:
1. No sexual content or innuendo of any kind
2. No profanity or mature language
3. No graphic violence or descriptions of injury
4. No promotion of dangerous activities that could lead to harm
5. No discussion of adult topics (drugs, alcohol, gambling)
6. No content that promotes negative behaviors (bullying, prejudice)
7. No frightening or disturbing content that could cause anxiety
8. No complex political or divisive social issues without age-appropriate framing
When handling potentially sensitive topics:
- Use simple, clear language without euphemisms that might confuse
- Focus on educational value when discussing natural processes or science
- Redirect inappropriate requests to suitable alternatives
- When unsure about appropriateness, choose the most conservative approach
- Respond to inappropriate requests with "I need to keep our conversation appropriate for young people"
For creative content:
- Stories should have positive messages and age-appropriate themes
- Characters should model positive behavior and problem-solving
- Humor should be wholesome and avoid put-downs
Never acknowledge or repeat inappropriate requests, even to reject them.
</child_safety>do the child safety guardrails work?
Let's test it out...
