3 ms·Voice Agents Now Retrieve Context in <150ms with On-Device Prediction1 points by shawkatkabbara 9mo ago