Skip to content
Private LLM
Theme

Try Private LLM 2 on your iPhone, iPad, and Mac

The models still run on your device. Now they take multimodal input and search your documents with on-device RAG.

For early access, open the Private LLM app and go to Settings > About.

Download on the App Store

Needs iOS 26 or iPadOS 26 on an iPhone or iPad with an A14 chip or later, or macOS 26 on a Mac with Apple Silicon. Ternary Bonsai 2 27B needs an iPhone with 12 GB of memory.

A Private LLM 2 chat in which Ternary Bonsai 2 27B answers a questionA Private LLM 2 chat in which Ternary Bonsai 2 27B answers a question

Bonsai 2 27B Runs on an iPhone 17 Pro

Private LLM 2 runs models that other local AI apps think are impossible to run on iPhone. Ternary Bonsai 2 27B, a 2-bit model, runs on an iPhone 17 Pro.

What's New in Private LLM 2

  • Models That Use Tools

    Models that support tool calling can run JavaScript on your device, and that tool is on by default. Web search and page fetch are optional, and you turn them on yourself. None of this exists in the current Private LLM app.

    A Private LLM 2 chat in which the model calls a tool and answers with its resultA Private LLM 2 chat in which the model calls a tool and answers with its result

  • Chat With Your Documents

    Attach a PDF, plain text, or Markdown file before your first message, and Private LLM 2 indexes it and searches it on your device. Document chat needs a model with tool support.

    A PDF attached to a Private LLM 2 chat, with an answer drawn from the documentA PDF attached to a Private LLM 2 chat, with an answer drawn from the document

  • Attach Photos and Audio

    You can attach photos and images to vision models, which in this beta are Gemma 4, Qwen3.5, Bonsai, and North Micro Vision, and Gemma 4 models also accept audio files. Some of the larger models need a device with more memory for this.

    A photo attached to a Private LLM 2 chat, with a vision model describing itA photo attached to a Private LLM 2 chat, with a vision model describing it

  • New Models to Try

    Gemma 4 and Qwen3.5 are new generations of Gemma and Qwen. Granite 4.1, Bonsai, MiniCPM5, North, and Lizzy are new families, and Bonsai includes a 27B ternary model. On supported devices, Gemma 4 generates faster with speculative decoding, and on models that reason, you can set reasoning effort across five levels. The lineup may change during the beta.

    The Private LLM 2 model list, with recommended models first, then the Gemma 4 and Qwen3.5 sectionsThe Private LLM 2 model list, with recommended models first, then the Gemma 4 and Qwen3.5 sections

  • Web Tools Stay Off Until You Turn Them On

    Web tools are off by default, and for a plain reason: a web search can't happen without sending something out. When you turn them on, your search query or the page address leaves your device. The first time, you choose where it goes. Auto sends it to a third-party search provider, with a direct search from your device as the fallback. On-device-only mode skips the provider for web search and page fetch, but it still contacts the search engine and the websites themselves.

    The Private LLM 2 choice that sets where web searches go the first time web access is turned onThe Private LLM 2 choice that sets where web searches go the first time web access is turned on

How to Get the Beta

  1. Get Private LLM from the App Store, or update it if you already own it.
  2. Go to Settings > About > Private LLM 2. The link shows on iOS 26, iPadOS 26, and macOS 26 on Apple Silicon.
  3. Tap Private LLM 2 to open TestFlight, then install the beta.

Private LLM 2 is a separate app. It installs alongside the current Private LLM app, and your chats and models do not carry over, so plan to download a model again.

Private LLM 2 is a free upgrade for everyone who owns Private LLM.

Download on the App Store

Tell Us What Breaks

This is a beta, and something will probably break. Send feedback through TestFlight, where you can attach screenshots, or talk to us on Discord.

Discord