Skip to content
HUIAHUB

Assistants & chat

KernelAI

An iOS and Android app that runs 53 open-weight models offline on your phone, with no account or subscription.

Verified Free

What KernelAI does

KernelAI is a mobile app that runs open-weight language models on the phone itself. The catalog has 53 models from 15 labs, including Qwen, Meta, Microsoft, Google and DeepSeek, ranging from 112 MB to 3.11 GB. Before you download a model, the app checks it against your device's memory and disables any model that will not fit, printing the required RAM next to what the phone has.

Models run as quantised files through llama.cpp. After a one-time download, prompts are processed on the device and the app works in airplane mode. You can attach a PDF, a Word file or a spreadsheet and ask about it, point the camera at a sign or a receipt, or ask for code with syntax highlighting. Dictation uses on-device speech recognition. A custom system prompt with presets sets the tone and format of every chat, and you can import your own GGUF file from Files or a URL as an experimental model.

Web search is off by default. Adding your own Tavily or Exa API key turns it on for models tagged Web Search, and answers then list their sources. Two Shortcuts actions, Ask KernelAI and Extract Text from Document, let automations and Siri use the local models.

The app needs iOS 16.4 or later on iPhone and iPad, or Android 7.0 or later. It suits people who want a private chat assistant that keeps conversations on the phone and runs without a network connection.

  • Chatting with an open-weight model offline on a phone
  • Asking questions about a PDF, Word file or spreadsheet without uploading it
  • Reading text from a photo of a sign, menu, chart or receipt
  • Running local prompts from iOS Shortcuts automations

Who should use it

  • People who want chats to stay on their own device
  • Users who need an assistant that works without a network connection

Who should look elsewhere

  • People who want hosted frontier models, since KernelAI runs only local open models of up to 3.11 GB
  • Devices below iOS 16.4 or Android 7.0

Strengths

  • Free, with no account, subscription or usage caps
  • Checks each model against the phone's memory before download
  • Runs offline after a model is downloaded
  • Supports importing your own GGUF models

Limitations

  • Web search needs your own Tavily or Exa API key
  • Imported GGUF models are marked experimental because compatibility varies by file
  • Which models run depends on the phone's memory; models that will not fit are disabled

Privacy notes

The privacy policy says the app does not sync conversations to a cloud service or to a server the developer runs. Model downloads come from Hugging Face, optional web search sends a short query to the chosen provider, and limited diagnostics go to Sentry.

Sources