Natural voices sound like a real person. They take about a second longer to start than device voices. If the free limit runs out, .thing switches to a device voice.
This device has no voices with that accent. Microsoft Edge includes extra online voices (including Nigerian and Canadian English), and on a Mac you can add more in System Settings, Accessibility, Spoken Content.
.thing admin
Moving up a Gemini tier
Open Google AI Studio and turn on billing for the project that owns .thing's key. The same key keeps working with the higher limits, so nothing here needs to change.
If you create a key in a different, paid project instead, paste it under AI service, press Test connection, then Save & apply.
Optionally switch to stronger models under AI service. Paid tiers unlock models the free tier doesn't include.
Under Data & publish, download the updated thing.html and replace the copy you share or host.
A Gemini key from Google AI Studio, or a Claude key (starts with sk-ant-).
When set, .thing talks to Gemini through your key server, and the key is left out of the published file so visitors can't see it. Setup steps are in the README that comes with the hosting package.
One per line, in order of preference. .thing uses the first one that works. Fast models feel best for voice.
Models available to this key
Add a model to a list with its buttons. Any listed model this key can't use is named above, next to Test connection.
Turn off to save your quota. .thing then uses the voices built into each device.
Instructions for everyday conversation. .thing's name and voice are added automatically.
Counts from this browser only. They help you see when the free limits start to bite.
Recent errors
Publish
Save & apply changes only this browser. To change .thing for everyone, download the updated file and replace the copy you share or host (index.html on your website).
