✦ Aetheris AI

One app. Every great AI model.

A beautiful, fast, fully native Android chat client for OpenAI, Anthropic, Groq, OpenRouter, and any OpenAI-compatible endpoint — with streaming responses, encrypted on-device API keys, and a polished Jetpack Compose UI.

---

✨ Why Aetheris

Most chat apps lock you into one provider. Aetheris lets you talk to every major LLM — switch models mid-conversation, bring your own keys, and keep everything encrypted on your device. No accounts, no servers, no telemetry.

⚡ Streaming-first · 🔐 Encrypted keys · 🎨 Designed, not assembled · 🪶 Single 100% Compose module

🚀 Features

• 🔌 Multi-provider — OpenAI, Anthropic (Claude), Groq, OpenRouter, plus unlimited user-defined custom providers (Ollama, Together, LM Studio, vLLM, your own gateway, …) — both OpenAI-compatible and Anthropic-compatible API styles
• 🔭 Live model discovery — fetch the model catalog from any provider via GET /v1/models, cached in Room; pin / hide / refresh per provider
• ➕ Add your own models — manually add any model id to any provider, in case its catalog endpoint is unavailable
• 💬 Real streaming — proper SSE framing for both OpenAI [DONE]-style and Anthropic typed events
• 🔐 Encrypted API keys — EncryptedSharedPreferences with MasterKey (AES‑256‑GCM); separate key per provider (built-in or custom); never backed up to the cloud, never sent anywhere except the provider you chose
• 💾 Persistent conversations — Room database, auto-titled from your first message
• 🧠 Per-conversation model + system prompt — change provider / model on the fly from the chat header
• 📝 Markdown rendering — code blocks, lists, links, strikethrough via Markwon
• 🎨 Aetheris design system — custom Material 3 theme (Electric Violet / Cyan / Amber), edge-to-edge, light + dark, follows your system or your preference
• ✋ Stop generation, ⏎ multi-line input, 📋 one-tap copy, ⏱ live typing indicator
• 🛡 Secure by default — release builds disable HTTP body logging, HTTPS-only network policy, allowBackup="false", data-extraction rules exclude the encrypted store and the chat DB

📸 Screens

Conversations	Chat (streaming)	Settings
Add screenshot	Add screenshot	Add screenshot

Drop screenshots into docs/ and reference them here once you have a build.

🧱 Architecture

A clean MVVM + repository split, single Gradle module, package root com.aetheris.chat.

ui/               ← Compose screens (chat, conversations, settings) + ViewModels
ui/components     ← ChatBubble, MessageInput, ModelSelector, TypingIndicator
ui/theme          ← Aetheris Material 3 palette, typography, theme
data/local        ← Room: AppDatabase, ConversationDao, MessageDao
data/remote       ← LlmClient (OkHttp + SSE) + OpenAI / Anthropic DTOs
data/repository   ← ChatRepository, SettingsRepository, ProvidersRepository, BackupRepository
di/               ← Hilt modules (AppModule, DatabaseModule)

┌──────────────┐   StateFlow   ┌────────────────┐
│  Compose UI  │ ◄──────────── │   ViewModels   │
└──────────────┘               └───────┬────────┘
                                       │
           ┌───────────┬───────────────┼───────────────┬───────────┐
           ▼           ▼               ▼               ▼           ▼
  ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
  │    Chat      │ │   Settings   │ │  Providers   │ │   Backup     │
  │  Repository  │ │  Repository  │ │  Repository  │ │  Repository  │
  └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘
         │                │                │                │
         ▼                ▼                ▼                ▼
   ┌──────────┐   ┌──────────────┐  ┌──────────┐    ┌──────────┐
   │  Room DB │   │  DataStore + │  │  Room DB │    │  File I/O│
   │ Conv/Msg │   │  Encrypted   │  │  Models  │    │  AES-256 │
   └──────────┘   │  SharedPrefs │  └──────────┘    └──────────┘
                  └──────┬───────┘
                         │
                         ▼
                  ┌──────────────┐
                  │   LlmClient  │
                  │  OkHttp+SSE  │
                  └──────────────┘

🛠 Tech Stack

Layer	Tools
Language / Build	Kotlin 2.1, AGP 8.7.3, Gradle 9.5, JDK 17
UI	Jetpack Compose (BOM 2024.12), Material 3, Navigation-Compose, Material Icons Extended
DI	Hilt 2.53 + hilt-navigation-compose
Persistence	Room 2.6, DataStore Preferences, EncryptedSharedPreferences (security-crypto 1.1.0)
Networking	OkHttp 4.12 (raw SSE), kotlinx-serialization-json
Markdown	Markwon 4.6 (core + strikethrough)
Async	Kotlin Coroutines + Flow
Min / Target SDK	26 / 35

📦 Supported providers

Provider	Type	Example models
OpenAI	OpenAI-compatible	GPT‑4o, GPT‑4o mini, GPT‑4.1 family, o4‑mini
Anthropic	Native /v1/messages	Claude Sonnet 4, Claude 3.5 Sonnet, Claude 3.5 Haiku
Groq	OpenAI-compatible	Llama 3.3 70B, Mixtral 8x7B, Gemma 2 9B
OpenRouter	OpenAI-compatible	GPT‑4o, Claude, Gemini 2.5 Pro, Llama 3.3 (via OR)
Custom (any number)	OpenAI- or Anthropic-compatible	Any base URL + model id (Ollama, vLLM, LM Studio, Together, your own gateway, …)

Models are seeded from a curated bootstrap list and then augmented at runtime by GET /v1/models — tap the refresh icon in the model picker (or in Settings) to pull the latest catalog from any provider. Manually-entered models persist alongside discovered ones; pin favorites to the top, hide noisy entries.

🏁 Getting Started

Prerequisites

• Android Studio Ladybug (2024.2) or newer
• JDK 17
• An Android device or emulator running API 26+

Build & run

git clone https://github.com/rahulmasal/AetherisAI.git
cd AetherisAI
# Open in Android Studio, then Run ▶ — or:
./gradlew assembleDebug
./gradlew installDebug

If gradlew is missing on a fresh checkout, run gradle wrapper once or open the project in Android Studio to generate it.

First launch

1. Open Settings → paste your API key for any built-in provider (saved encrypted on-device).
2. Optionally tap Add custom provider to register an Ollama / vLLM / LM Studio / Together / private gateway endpoint. Pick the API style (OpenAI- or Anthropic-compatible), set the base URL, and save its API key.
3. Tap Fetch models on a provider card to populate the model catalog from GET /v1/models. Or use Add model to enter ids manually.
4. Pick your provider + model from the chat header (or hit the refresh icon there). Start chatting. Tap ⏹ to stop streaming, ✦ for a new chat.

🔐 Privacy & security

• API keys live in EncryptedSharedPreferences (AES‑256‑GCM keys / AES‑256‑SIV names) backed by Android's hardware-backed MasterKey.
• HTTP request/response bodies are logged only in debug builds — release builds never write your keys to logcat.
• network_security_config enforces HTTPS-only by default.
• allowBackup="false" and data_extraction_rules.xml keep the encrypted prefs, the Room DB, and DataStore out of cloud backups and device-transfer flows.
• No analytics. No crash reporting. No remote config. Your messages go from your phone straight to the provider you picked.

🗺 Roadmap

• Image input (vision models)
• Message editing & regeneration
• Conversation export (Markdown / JSON)
• Search across conversations
• Custom theme accents
• Tablet / foldable layouts
• Voice input
• Local model support via llama.cpp / MLC-LLM

🤝 Contributing

Issues and PRs are very welcome. If you're adding a feature, please:

1. Open an issue first to discuss the approach.
2. Match the existing Compose / MVVM style.
3. Keep dependencies lean — Aetheris ships small on purpose.

📜 License

MIT © Aetheris AI contributors

---

Crafted with care for people who love good software and good AI. ✦