Jack Dorsey’s Block released Dessa, a full-duplex speech-to-speech model from its research lab that runs locally on consumer hardware (MacBook M4 Max or better) and was trained on human speech using a single NVIDIA GPU rack.
Dessa can talk and listen simultaneously (like OpenAI GPT-Live). Block argues text-in-the-middle pipelines discard emotional cues; Dessa embeds emotion (laughs, pauses, tired/crying cadence) without ElevenLabs-style voice tags.
Knowledge comes from conversational training data rather than pairing with a separate LLM for retrieval. No public access yet; Block invited safety researchers to test deception/social-engineering risks.
Square already uses voice AI for ordering; on-device Dessa could lower cost for merchant and consumer experiences. Jack Dorsey promoted demos on X Oct 23, 2026.