Apple Silicon Macs running macOS 26 Tahoe ship with a built-in ~3B parameter language model (Apple Foundation Model) as part of Apple Intelligence. A third-party tool called Apfel exposes this hidden model as a local OpenAI-compatible server on port 11434, making it usable by any client that speaks that protocol — including Home Assistant via the Extended OpenAI Conversation HACS integration. The model has significant limitations: a 4,096-token context window, weak factual recall, poor math, and no complex code generation. To make it work with Home Assistant, tools and function calls must be disabled to stay under the token limit. While it won't replace cloud assistants or most self-hosted local LLMs, it's a free, fully on-device option already sitting on every modern Mac.

4m read timeFrom xda-developers.com
Post cover image
Table of contents
Apple already put a language model on your MacApfel turns that model into a server anything can talk toThe 4,096-context window limits what it can doIt doesn't have to beat the cloud
79 Impressions