CoreAI Zoo vs Sandbox Relay

CoreAI Zoo puts the open-weight AI model zoo on your iPhone, iPad and Mac. Pick a model, download it once, and run it entirely on your device. No account, no server, and nothing you type, say or photograph ever leaves the device. CHAT Stream answers from open language models such as Qwen3, Gemma, LFM2.5, Granite and Mistral, from 0.6B to 30B+ parameters depending on your device's memory. Vision-language models (Qwen3-VL, MiniCPM-V, LFM2.5-VL) answer questions about your photos. VISION Ask questions about an image, read text with on-device OCR, and describe what the camera sees. AUDIO Transcribe speech with Whisper large-v3-turbo, Parakeet and Qwen3-ASR. Talk to an audio-understanding model, generate speech with Kokoro and VoxCPM, and try on-device music generation. CAMERA Live depth estimation, object detection (YOLOX, RF-DETR), action recognition and super-resolution on the camera feed. TOOLS Experimental models such as diffusion-based language models. BENCH Measure load time, tokens per second and memory for any chat model on your own hardware. HOW IT WORKS Every model runs through Core ML on Apple silicon. Weights are downloaded from the Hugging Face Hub on first use over HTTPS, then cached, so the app works offline afterwards. One model is resident at a time to stay within memory limits. New models arrive through a live catalog without an app update. GETTING STARTED Start with Qwen3-0.6B (454 MB) for the fastest first run. Larger models state the free space and memory they need; a recent iPhone Pro or an Apple silicon Mac runs the mid-size ones comfortably. Model weights follow their own licenses, listed on each model's Hugging Face page. The app itself is BSD-3-Clause and part of the open CoreAI model zoo. Report issues and request models at github.com/john-rocky/coreai-model-zoo.

For developers who run Claude Code or Codex in Docker Sandbox, Sandbox Relay is the phone client for driving that agent. It is a companion, not a standalone app: it needs a broker you run, with an agent connected behind it. Kick off a task from the train, check on a long-running job from the couch, or steer the agent while you are away from your desk. Docker Sandbox runs each agent in its own isolated micro VM. A small broker you self-host sits in the middle: you send work from your phone, the broker passes it to the agent, and the streamed reply comes back. Your infrastructure, your agent. The app is just the remote. New to this setup? The app's support page explains how it works and what you need to run it. WHAT YOU CAN DO - Connect to your broker with a URL and a bearer token. Connections use HTTPS, with plain HTTP allowed only for local-network addresses. - See every connected agent at a glance, with live online/offline status. - Send the agent work and watch its reply stream in token by token. Stop a turn mid-stream, or start a fresh thread whenever you want. - When an agent goes offline, its thread stays readable as history. - See per-turn input and output token counts when the broker reports them. - Save reusable prompts as local notes, organised by hierarchical tags like #work/project/feature. Drill down by tag, copy a note straight into a thread, or link a note to the thread it came from. Notes live on your device only; they never touch the broker. WHAT YOU NEED - A broker you run, reachable from your phone (over your local network, or remotely through a tunnel or private network you control). - A coding agent, Claude Code or Codex, running in Docker Sandbox and connected to that broker. PRIVACY Sandbox Relay talks only to the broker you configure. Your access token is stored securely on your device and sent only to that broker, to authenticate you; it never goes to the developer or any third party. Your notes stay on your device and are never sent anywhere. No third-party analytics, no account to create. Note that the app streams whatever your agent outputs, verbatim.