Today we are launching telli Duo agents.
Over the past years, the industry standard for voice AI has been the cascading architecture.
Meaning when a human speaks to an agent, the voice gets transcribed by an STT (speech-to-text) model, goes into an LLM, and is transcribed back with a TTS (text-to-speech) model.
This worked well and kicked off the era of Voice agents. But it came with fundamental flaws: It's impossible for an agent to listen and speak at the same time.
A good conversation has a rhythm. We pause, interrupt, and respond before the other person has finished their sentence. It happens so naturally that we rarely think about it. Building this into an agent was impossible.
A couple of months ago we got early access to @OpenAI's GPT-live model. It made us rethink our entire Voice AI infrastructure and rework it to bet on full-duplex speech-to-speech models. The result was the new telli Duo agent.
We went deeper into how we did that in our blog post: hi.telli.com/duo-agents
PS: We know we over-exaggerate the difference in the video. The comparison is more with a traditional IVR bot. But it's still much better now, just harder to show in a demo than when you experience it yourself. Just give yourself a call on telli.com.



