Repolode
Submit
WWRJ3321

WWRJ3321/IRIDIS

An Android app that runs LLMs fully offline on-device, with local agents, memory, voice and an OpenAI/Ollama-compatible LAN API.

View on GitHub
Stars
95
Forks
8
Open issues
3
Last commit
17 hours ago

What is it?

IRIDIS is an all-in-one on-device AI platform for Android. It runs language models directly on the phone through three inference engines: llama.cpp (GGUF models), MNN-LLM, and LiteRT-LM, which can use CPU, GPU, or a Qualcomm NPU depending on the device. Chat, speech, and conversation records stay on the device, and the app works in airplane mode.

Beyond chat, the app includes an agent mode that can read and write files, search the web, and run scheduled tasks, with changes that can be undone. It also offers long-term memory, local document retrieval, offline speech recognition and synthesis, and hands-free voice conversation. When enabled in settings, the phone can serve a local API compatible with OpenAI and Ollama on the LAN, requiring an API key for non-local access.

The project requires Android 12 or later on arm64 and is licensed under Apache-2.0. Building from source needs JDK 17, Android SDK 35, and NDK r27c or later.

Key features

  • 01Runs GGUF, MNN and LiteRT-LM models on-device with CPU, GPU or Qualcomm NPU support.
  • 02Works offline, keeping chats, speech and records on the phone.
  • 03Agent mode can read and write files, search the web and run scheduled tasks, with undoable changes.
  • 04Long-term memory and local document retrieval.
  • 05Offline speech recognition and synthesis with hands-free conversation.
  • 06Local API compatible with OpenAI and Ollama formats, reachable on the LAN with an API key.

Use cases

  • Chatting with an LLM on an Android phone without sending data to the cloud.
  • Querying private local documents with a model that never leaves the device.
  • Using the phone as a local model server for other devices on the same network.
  • Running an offline voice assistant for hands-free use.

Quick start

  1. 1

    Download and install the APK from the Releases page

  2. 2

    Download or import a model in Model Management

  3. 3

    Start chatting; open Agents when you need tools

Things to know

  • Requires Android 12+ on arm64 (Snapdragon 8 Gen 1 or newer), so it does not run on other platforms.
  • Models with 4B or more parameters are recommended to have 12 GB of RAM.
  • Some optional voice models are limited to non-commercial use.
  • The Qualcomm QNN runtime for NPU is proprietary and not distributed with the source code.

In collections

Run LLMs locallyPrivacy & local-first toolsAndroid apps & phone tools

Topics

#android#on-device-ai#llm#offline#local-llm#llama-cpp#privacy#voice-assistant

FAQ

What is IRIDIS?
An Android app that runs LLMs fully offline on-device, with local agents, memory, voice and an OpenAI/Ollama-compatible LAN API. IRIDIS is an all-in-one on-device AI platform for Android. It runs language models directly on the phone through three inference engines: llama.cpp (GGUF models), MNN-LLM, and LiteRT-LM, which can use CPU, GPU, or a Qualcomm NPU depending on the device. Chat, speech, and conversation records stay on the device, and the app works in airplane mode.
Is IRIDIS free to use? What license does it have?
IRIDIS is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is IRIDIS still maintained?
Actively maintained. The last commit to IRIDIS was on Nov 6, 2026.
What can I use IRIDIS for?
Chatting with an LLM on an Android phone without sending data to the cloud. Querying private local documents with a model that never leaves the device. Using the phone as a local model server for other devices on the same network. Running an offline voice assistant for hands-free use.
How do I install IRIDIS?
Download and install the APK from the Releases page Download or import a model in Model Management Start chatting; open Agents when you need tools
Where can I get IRIDIS?
The source code and installation instructions are on GitHub: https://github.com/WWRJ3321/IRIDIS

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects