How it works
Tap the mic, ask a question out loud, and KidQA reads back a short answer with a picture. The AI is tuned for ages 3–6, so answers stay to a sentence or two in words a young child knows.
Where the pictures come from
Each answer comes with a short, self-contained search phrase (e.g. “blue whale swimming ocean”). KidQA sends that phrase to Pixabay, a free stock-photo library, with safe-search turned on. We also run our own keyword filter on top — any result tagged with weapons, violence, drugs, or other adult topics is skipped. If nothing safe matches, no picture is shown.
Safety
KidQA is always in kid mode — there are no settings to unlock and no adult mode hiding behind a toggle. The AI is instructed to avoid violence, scary content, and adult topics, and to gently redirect unsafe requests (“let’s ask a grown-up about that one”) instead of refusing or going dark.
The three modes
KidQA has three ways to ask a question. They all give the same kind of answer card — the difference is which voice pipeline runs underneath.
- Realtime (default, at /) — Uses OpenAI’s Realtime voice API over WebRTC. Voice goes in and voice comes out directly, so the back-and-forth feels like a conversation. Fastest and most natural, best on good Wi-Fi.
- Standard (/standard) — The browser transcribes speech to text, the text goes to OpenAI for an answer, and a separate text-to-speech step reads it back. More steps means a small pause before the answer, but it works reliably on more browsers and weaker connections.
- xAI (/xai)— Same realtime style as the default, but powered by xAI’s Grok voice instead of OpenAI’s. Useful for comparing how a different voice and model handle the same kid’s question.
Question history is saved in your browser’s local storage for the history page. In Standard mode, KidQA may include a few recent Q&A pairs with a new question so follow-ups make sense.