privateSLM 2.0: a chat app that uses the model already inside your phone

July 3, 2026 · 4 min read

Most local-AI apps start the same way: pick a model from a long list, download a couple of gigabytes, wait, and hope you chose well. We think that first experience is exactly backwards — especially now that your device probably ships with a very good small model already installed.

privateSLM 2.0 is a complete rebuild around one idea: use the best brain the device already has.

Zero-download by default

On devices with Apple Intelligence (iPhone, iPad, Mac), privateSLM talks to Apple's built-in ~3B model through the Foundation Models framework. Install the app, open it, type. No model list, no gigabytes, no setup. It runs on the Neural Engine and never touches the network.

No Apple Intelligence? The built-in model manager downloads a small open model — Llama 3.2 1B, Qwen2.5 1.5B or Gemma 2 2B — once, and runs it locally through llama.cpp. The engine switch is automatic and visible in the Models screen.

A real chat app, not a tech demo

  • Chat history that actually exists. Multiple conversations, search across titles and content, rename, delete. Your chats survive a restart — table stakes that local-AI apps somehow keep skipping.
  • Voice both ways. Dictate a question, hear the answer read aloud — a hands-free loop, fully offline.
  • Control over the generation. Edit any message and regenerate from that point. Stop mid-reply. Copy anything. Tune the system prompt, temperature and sampling.

Privacy as a property, not a promise

There is no account, no analytics, no server component at all. The airplane-mode test passes: everything above works with the radio off. Chats are stored only in the app's local database on your device.

One price

privateSLM is a one-time purchase — $4.99, family sharing included. No subscription, because on-device inference costs us nothing per message, so it shouldn't cost you either.

Get privateSLM on the App Store

Curious how the Foundation Models integration works under the hood — including the streaming bug we caught the day before review? Read Apple Intelligence for developers.