Just released v0.9.1 of afm, a CLI that exposes Apple's on-device Foundation Models through OpenAI-compatible API endpoints.

What's new in v0.9.1 - Built-in Web UI:

Links:

GitHub: https://github.com/scouzi1966/maclocal-api

Release: https://github.com/scouzi1966/maclocal-api/releases/tag/v0.9...

You can now run afm -w to start both the API server and a chat web interface in one command. It integrates llama.cpp's webui and auto-opens your browser. No need to set up Open-webui separately if you just want a quick chat interface.

afm -w

That's it. Browser opens to http://localhost:9999 with a chat UI talking to Apple's on-device 3B model.

Other changes:

/props endpoint for webui compatibility

model field now optional in chat completion requests

llama.cpp pinned as a submodule for reproducible builds

What afm does:

Runs Apple's 3B parameter on-device LLM

OpenAI-compatible API (/v1/chat/completions, /v1/models)

Single-prompt mode: afm -s "your question"

Pipe mode: cat file.txt | afm

LoRA adapter support for fine-tuned models

Vision capabilities (text extraction, table OCR)

Works as a backend for Open-webui too

Install:

brew tap scouzi1966/afm

brew install afm

Requirements:

macOS 26 (Tahoe)

Apple Silicon (M1/M2/M3/M4)

Apple Intelligence enabled

Questions welcome.