Introducing Duo

Duo is a full-duplex conversational AI. It listens while it speaks and hears your voice directly instead of a transcript, so it reacts to pauses, a quick "mm-hm" and interruptions as they happen.

Rather than training a speech model from scratch, we start from an open-weight language model and adapt it to real-time conversation. Duo keeps the model's knowledge, intelligence and instruction following, and as open-weight models improve, Duo can improve with them.

We're preparing an open-weight release of our preview model, which runs on a laptop, together with a technical write-up of how we built it.

We want talking to AI to feel as natural as talking to a person. Not so AI can pretend to be human, but so we can be more human when we use it.

We're looking to partner with companies to build exceptional conversational interfaces tailored to their products, with their own voice, domain knowledge and way of talking to customers. We take care of the data, training and evaluation needed to adapt Duo to each product.