A Show HN thread on Hacker News is currently exploding — 628 points and 137 comments in a short time. The project is called Apfel, and the premise is simple enough to make one wonder why no one has done it before: Apple has built a full-fledged language model into macOS 26 Tahoe as part of Apple Intelligence, but they expose it almost exclusively through Siri and their own system apps. Apfel breaks down that wall.
Specifically, this means you can pipe text through the terminal, send files to the model, or spin up a local HTTP server with an OpenAI-compatible API — and use existing tools and SDKs as if you were talking to GPT-4. The difference is that everything happens on your own machine, without a single byte leaving it.
The model has a context window of 4096 tokens and supports tool calling, which is more than enough for many practical applications. It's not GPT-4o, but that's not the point either. The point is that it exists, it's fast enough, it's private, and it costs nothing.
What distinguishes Apfel from Ollama or LM Studio? Quite a lot, actually. Ollama and LM Studio are both brilliant tools, but they are about downloading third-party open-source models and running them locally. That requires time, disk space, and a bit of technical comfort. Apfel does neither — the model is already there. It's closer to discovering that your laptop has a hidden turbo button.
For developers, the HTTP server with an OpenAI-compatible API is the real killer feature. It means you can replace the API call in existing projects with a single line change and suddenly have zero cost, zero latency (network-wise), and full privacy — on machines that are already running macOS 26.
Of course, there are limitations. You need Apple Silicon and the latest macOS, and Apple's model is not open in the traditional sense — you cannot fine-tune it or replace it. And right now, these are early signals from community sources on HN; we don't yet know how the model performs on more demanding tasks over time.
But the hype factor is real: people in the thread are surprised that this hasn't received more attention. Many are discovering for the first time that they own hardware with a built-in LLM they've never used.
Worth keeping an eye on. This could become one of those quiet things that are suddenly everywhere.