Fully on-device.
Chat, vision, voice, and image generation run locally on Apple silicon. No round trip to a server for a normal request.
Run AI on your iPhone.
Not in someone else's cloud.
Remote ARM runs chat, vision, voice, and image models entirely on-device — then, if you flip it on, serves them as an authenticated API on your local network.
OpenAI-compatible·
Anthropic-compatible·
Ollama-compatible·
Bearer-token auth·
curl-friendly·
on-device inference·
OpenAI-compatible·
Anthropic-compatible·
Ollama-compatible·
Bearer-token auth·
curl-friendly·
on-device inference·
Local-first meets developer-ready.
Nothing leaves your phone unless you ask it to.
OpenAI, Anthropic & Ollama shaped.
Point an existing client at your phone. /v1/chat/completions, /v1/messages, /api/chat — no adapter needed.
Authenticated by design.
Remote Node mode is opt-in and bearer-token gated, with an HTTPS option pinned by SHA-256 fingerprint.
One queue per resource.
Chat and image generation share an admission queue; embeddings, transcription, and speech share another.
The round trip is over.
No cloud in between. No per-token bill.
Local model catalog.
Discover, download, import, and manage compatible models right from the device.
Device-aware model fit.
Real memory-ceiling checks before you load a model your hardware can't hold.
Camera, voice & image.
Vision, dictation, and on-device image generation, where the model and device support them.
Retrieval & tools.
Local document indexing and tool-use workflows, without uploading files anywhere.
Local inference stays on the device. Nothing is uploaded to train a model.
Talk to it like an API.
Remote ARM speaks the shapes your tools already know — OpenAI's Chat Completions and Responses APIs, Anthropic's Messages API, and the Ollama API. Point any compatible client at your phone's local address and it just works.
/v1/chat/completions
/v1/messages
/api/chat
/v1/audio/speech
/v1/images/generations
# from any machine on the same network
curl http://remote-arm.local/v1/chat/completions \
-H "Authorization: Bearer <token>" \
-H "Content-Type: application/json" \
-d '{"model":"local","messages":[{"role":"user","content":"Hello"}]}'Build it yourself.
Open source, no App Store gate. Sideload it your way.
Xcode
Clone the repo and build straight from source with XcodeGen.
- Full source, MIT-style licensing
- XcodeGen-managed project
- Runs great in Simulator
Sideload
Build once, install with your own Apple ID or developer certificate.
- Bring your own signing
- Works on real hardware for model-fit & camera
- Full control, no third-party store
FAQ.
For the ones running local.
Does any of my data leave the device?
No. Inference runs on-device. Network activity only happens for things you explicitly do: searching for a model, downloading one, or turning on Remote Node mode.
What is Remote Node mode?
An optional, bearer-token-authenticated HTTP API that exposes your device's local model to other computers on the same local network. It's off by default and never exposed publicly.
Which clients can I point at it?
Anything that speaks the OpenAI, Anthropic, or Ollama HTTP API shapes. Set the base URL to your device's local address and the bearer token shown in Settings.
Is Remote ARM on the App Store?
Not currently. It ships as something you sideload yourself, or build from source with Xcode.
What do I need to run it?
An iPhone or iPad on iOS 18 or later. Which models you can run — and how fast — depends on the device's memory and chip.
Your phone. Your model. Your rules.
View on GitHub