Lore

Local Research for Android

An AI assistant that runs on your phone.

Lore chats, researches the web, remembers what matters to you and reads your files, while its AI model runs entirely on your phone, with no accounts, no Lore servers, no analytics and no ads.

How it works Privacy policy

Coming soon to Google Play. Needs Android 15 or newer and about 8 GB of RAM.

How it works

Lore runs Gemma 4, an open AI model from Google, on your phone with LiteRT-LM. Your questions are answered there, not by an AI service in the cloud.

  1. Download the model once

    On first launch, Lore downloads its model (about 2.6–3.7 GB, depending on your phone). After that it lives on your phone.

  2. Ask anything

    The model on your phone reads your message, thinks it through and writes the answer. You can watch each step as it happens.

  3. The web, only when needed

    When a question needs current facts, Lore searches the web and reads the pages itself, then answers with links to its sources.

What Lore can do

A familiar chat, with an assistant that does its own legwork.

Chat

Ask questions, think things through, draft and rewrite. Lore shows its thinking and every step it takes, and keeps your Chats in a side drawer.

Web research

Lore searches with DuckDuckGo, or Brave Search if you add your own API key, reads the pages it finds and cites them as links you can check.

Memories

Lore saves lasting facts you tell it, such as your preferences, as short notes that every Chat can use. View, edit or delete them on the Memory screen.

Attachments

Add photos, PDFs with text, and text files such as notes, CSV, JSON or code. Lore reads them on your phone and answers questions about them.

Runs in the background

Start a long answer and switch apps. Lore keeps working with an ongoing notification you can pause from, and carries on where it left off if it's interrupted.

Deep research Experimental

Lore drafts a research plan for you to approve, looks into each part and writes a report with its sources. It's experimental: small on-device models can get details wrong, so check the sources.

What leaves your phone

Only what your request needs, and only when it needs it.

  • Web search queries Sent to DuckDuckGo, or to Brave Search when you've added your own Brave API key.
  • Page requests Sent to the websites Lore reads to answer you, as a browser would.
  • A one-time model download About 2.6–3.7 GB, from models.rohittp.com (Cloudflare R2), with Hugging Face as the fallback.

No accounts, no Lore servers, no analytics and no ads. Lore has no servers of its own, so your Chats, Memories and Attachments are never sent to us.

Crash data reaches us only through Google Play's Android vitals, if you've allowed your phone to share usage and diagnostics. “Share diagnostics” in the app is a file you send yourself, if you choose to. Read the privacy policy.

Requirements

Running a capable AI model on a phone takes a recent, well-equipped device.

Android 15+
Android 15 or newer.
8 GB RAM
About 8 GB of memory or more. Phones with 12 GB get the larger model.
Supported GPU
A GPU that LiteRT supports, for answers at a usable speed.
Free storage
Room for the model (about 2.6–3.7 GB) plus 1 GB.

Phones with less memory see an “Unsupported device” screen, and Lore downloads nothing. Lore needs the internet for the model download and for web search. See the FAQ.