Show HN: Aotol AI – Offline LLM app runs on iOS with voice and multilingual (apps.apple.com)

Aotol AI is a new iOS app that runs a fully local large language model on-device, offering text and real‑time voice chat with multilingual support (English, Japanese, Simplified Chinese) and the ability to switch languages on the fly — all without an internet connection. The app bundles a ~1.9 GB footprint and requires iOS/iPadOS 17.6+, macOS 14.6+ on Apple M1 or later, or visionOS 1.3+, and advertises no accounts, no tracking and “100% offline” operation. Aotol AI Plus ($4.99) unlocks voice chat and advanced features; version 1.7 adds on‑device memory and Retrieval‑Augmented Generation (RAG) support for smarter, context‑aware conversations. For the AI/ML community this highlights the push toward practical on‑device LLM deployments: low latency, better privacy, and offline availability at the cost of shipping models and relying on mobile compute (likely quantization, model distillation, and Apple Neural Engine acceleration). On‑device RAG implies local embedding/index storage and retrieval pipelines, and real‑time voice implies local speech recognition/TTS or lightweight audio models. The app’s privacy claims are notable but worth auditing (store size and bundled model indicate tradeoffs in capability vs. device resources), and it’s another sign that edge LLM experiences are becoming increasingly viable for consumer applications.