Volocal

Talk to an AI assistant that runs entirely on your iPhone. No cloud servers, no API keys, no subscriptions -- just real-time voice conversations powered by on-device intelligence. FULLY PRIVATE, FULLY LOCAL Every word you say stays on your device. Volocal runs three specialized AI models directly on your iPhone's hardware -- nothing is ever sent to the internet. After downloading the models on first launch, you never need a network connection again. REAL-TIME VOICE CONVERSATIONS Speak naturally and hear responses streamed back in real-time. Volocal uses a three-stage pipeline -- Speech-to-Text, Language Model, and Text-to-Speech -- all running simultaneously on different chips inside your phone for smooth, low-latency conversations. INTERRUPT ANYTIME Change your mind mid-sentence? Just start talking. Volocal supports barge-in with hardware echo cancellation, so the microphone stays active even while the assistant is speaking. No need to wait for it to finish. NO SUBSCRIPTIONS, NO API KEYS Volocal is completely free and open source. There are no in-app purchases, no subscriptions, and no hidden costs. Download the models once (~2.3 GB, Wi-Fi recommended) and you're set. HOW IT WORKS Volocal splits the workload across your iPhone's specialized processors: - Speech recognition (Parakeet EOU) runs on the Neural Engine - Language model (Qwen3.5-2B) runs on the GPU via Metal - Text-to-speech (PocketTTS) runs on the Neural Engine This separation means all three models can run at the same time without competing for resources, keeping conversations fluid and responsive. LIGHTWEIGHT Total runtime memory usage is approximately 1.2 GB, well within the limits of modern iPhones. REQUIREMENTS - iOS 17 or later - Physical device only (Neural Engine is not available in the Simulator) - ~2.3 GB of storage for model downloads on first launch Volocal is MIT licensed and open source. Built by Fikri Karim.