Pocket AI Lab

A local AI app for iPhone that checks model fit before download and supports offline chat, vision input and on-device Whisper transcription.

Pocket AI Lab runs AI models on an iPhone and helps you judge which ones the phone can handle. It's for people who want private chat on their device and want to check memory and storage before downloading model weights.

Model fit checks: Get recommendations based on the memory iOS actually allows the app to use, available storage and the selected model's quantization.
Backend choice: Chat through MLX, llama.cpp for GGUF models, Core ML or Apple Intelligence, with streaming responses and conversation history across the supported backends.
Images and speech: Use vision models with images and video frames, or transcribe speech on the device with Whisper.
Hugging Face imports: Import models by repository link, with support for GGUF, safetensors, Core ML and Whisper weights and a fit check before downloading.
Live device readings: See CPU use, memory use and thermal pressure while a model generates responses.

Chat works offline after model downloads, and conversations stay on the device. The app doesn't require an account or a server for inference. It uses internet access for Hugging Face downloads and metadata, plus model catalog updates from GitHub.

The stated requirements are iOS 26 or later and an iPhone 15 or newer, with at least 6 GB of RAM for local models. Individual models still need to fit the device's memory allowance. The app is open source under the MIT license.