We just gave AI the ability to know exactly when to speak. No wakewords. No cameras. 100% local. Yesterday, two voice agents sat in a meeting on our laptops, catching and answering questions in real-time. They’re not 'human-level' remote colleagues quite yet (some OpenAI API latency), but its a start. Our tiny Selective Auditory Attention (SAA) engine drops into any device—even smart glasses—plugs into any LLM with one Python line, and helps voice agents follow real-life group conversations. Conversation capabilities • 97% accuracy picking the right speaker(s) within 5 meters • Waits its turn, speaks at the right time. • Ignore irrelevant prompts—no false triggers. Engine specs • 10x smaller & faster than multimodal SOTA • Only 400k params, processing stays on-device • Works with OpenAI, Gemini, or your own model We're excited to participate in this next wave of natural human-machine collaboration. Join the wait-list for the SDK (first comment). Tag someone who could use this to build the next killer voice agent. 👇 #VoiceAI #EdgeAI #LLM #FutureOfWork #AudioTech