Maid – Mobile Artificial Intelligence Distribution

What it is
Maid is an open‑source Android app (built with React Native) that lets you talk to large language models (LLMs) from your phone. It can run models locally using the llama.cpp engine and GGUF files, or connect to dozens of remote inference providers (OpenAI, Anthropic, Ollama, LiteLLM‑proxy, etc.) via API keys.

---

Core capabilities

Feature	What you get
Local inference	Load any GGUF model (e.g., Qwen, Phi, TinyLlama) and generate text entirely on‑device – no internet needed.
Remote providers	One‑click configuration for services such as API Route, Anthropic, DeepSeek, Mistral, Novita, Ollama, llmman, OrcaRouter, LiteLLM, OpenAI, etc.
Model marketplace	Browse curated Hugging Face models inside the app and download them with a single tap.
Conversation management	Create, rename, delete chats; export/import histories as JSON.
Parameter tuning	Adjust temperature, top‑p, top‑k, context length and other generation knobs per session.
System prompt & persona	Set a global system prompt to give the assistant a consistent character.
Voice output	Pair with the companion Maise app for text‑to‑speech.
Sync via Supabase	Optional account login to back up settings and chat logs.
Material You theming	Light/dark themes that follow Android system settings.
Open source, no ads	MIT‑licensed, telemetry‑free.

---

How it works

1. Installation – Grab the APK from the GitHub releases badge or install from Google Play.
2. Choose a backend –
   • Local: Place a .gguf file on the device, select it, and Maid launches llama.cpp under the hood.
   • Remote: Enter an API key for any supported provider; Maid builds the request payload and streams the response.
3. Chat – Type or paste prompts, tweak generation parameters, and view the assistant’s replies. Chats can be saved, exported, or synced.
4. Optional TTS – Open Maise to have the response spoken aloud.

---

Getting started from source

# Clone the repo
git clone https://github.com/Mobile-Artificial-Intelligence/maid.git
cd maid

# Install JS dependencies
yarn install

# Run the test suite
yarn test

# Build an Android APK
yarn build-android   # → android/app/build/outputs/apk/release/*.apk

The repository also ships a PDF user manual in the latest release.

---

Who might use it?

• Mobile‑first AI enthusiasts who want an LLM on their phone without relying on cloud services.
• Developers looking for a reference React‑Native implementation of LLM inference and multi‑provider API handling.
• Researchers or hobbyists who need a quick way to test different GGUF models on Android hardware.

---

Limitations & caveats

• Local inference is limited by the device’s CPU/GPU power and the size of the GGUF model you can store.
• Remote usage still requires an external API key and incurs the provider’s cost/latency.
• The app does not include built‑in TTS; you need the separate Maise companion.

---

License & provenance

• MIT license, no telemetry, no ads.
• Not officially affiliated with Hugging Face, Meta, OpenAI, or any listed model provider.

---

Bottom line – Maid turns an Android phone into a portable LLM client, supporting both on‑device GGUF models via llama.cpp and a wide range of cloud APIs, all wrapped in a clean, themable UI.