OnDevice Local AI Studio Support
OnDevice Local AI Studio is a local-first AI workbench for iPhone. Run open-source language and vision models — Qwen, Llama, Phi, Gemma, SmolVLM — entirely on-device via Apple's MLX framework and llama.cpp. Live camera captions, hands-free voice chat, a code assistant, OCR, document scan, RAG, and authenticated OpenAI-/Anthropic-/Ollama-compatible local APIs. No accounts, no cloud inference by default, no telemetry.
Support
Common Topics
No. Every model runs on the device. The only network traffic is when you explicitly search or download a model from HuggingFace, and a live indicator shows every outbound request.
iOS 18 or later. iPhone 12 or newer is recommended. The app auto-picks a model that fits your device — 1.5B parameters on older iPhones, up to 7B on Pro Max devices.
No. There is no sign-up, no account, no phone number, and no telemetry. Open the app and use it.
The initial model download is around 1.5 GB minimum, depending on which model you choose. Wi-Fi is recommended for the first download.
A feature that pairs your Mac with your iPhone over the local network so the Mac can send inference requests to the phone. Everything stays on your own devices.
OnDevice Local AI Studio is the full workbench (UI + optional local APIs). OnDevice Local API Server (coming soon) and OnDevice CoreAI Local API Server (beta) are focused API-server products in the same OnDevice line.