Blog
Notes from inside the latency budget.
Engineering and product writing about what it actually takes to keep a conversation with an avatar under a second.
Engineering
August 4, 2026
Anatomy of a 790 ms reply
Where the time actually goes between a caller finishing a sentence and an avatar starting to answer, and which parts are worth fighting for.
7 min read
Product
June 19, 2026
Why we made every model swappable
The case for treating the language model, the recognizer and the voice as three replaceable parts rather than one bundled product.
5 min read
Engineering
April 28, 2026
Giving an avatar real tools
Tool calling in a live voice conversation is a latency problem before it is a capability problem. Here is how we cover the gap.
6 min read