ARIA is a fully local, privacy-first AI assistant designed for ultra-low latency interaction. Unlike cloud-based assistants, ARIA runs entirely on your hardware, ensuring zero telemetry and instant response times.
ARIA is built to be a portable AI system. Whether running from your PC or a USB drive, it provides a consistent, high-performance interface for voice-to-voice and text-to-voice interactions.
- 100% Local: No data leaves your machine.
- Modular Architecture: Easily extendable tool system.
- Hybrid Routing: Intelligent "Fast Path" for system commands, bypassing the LLM when not needed.
- Real-time Voice Input: Powered by
faster-whisperwith dynamic RMS silence detection. - Synchronized TTS: Uses
Piper TTSfor high-quality, ultra-realistic voice models, perfectly synchronized word-by-word with the terminal output.
- Intelligent Routing: Uses
rapidfuzzto launch local apps and execute system commands instantly. - Tool Execution: Native support for browser actions, file operations, and shell commands with strict schema validation.
- State Management: Thread-safe architecture prevents overlaps between listening, thinking, and speaking.
- Context Awareness: Short-term interaction buffers and long-term user preference storage.
- Dynamic Personalization: Remembers your name and preferences across sessions.
| Component | Technology |
|---|---|
| LLM Engine | Ollama (Default: phi3) |
| STT | faster-whisper (Base.en) |
| TTS | Piper TTS (en_US-lessac-medium) |
| Interface | prompt_toolkit |
| Audio | pygame & pyaudio |
| Fuzzy Matching | rapidfuzz |
- OS: Windows 10/11 (Required for Piper
.execompatibility) - Python: 3.10 or higher
- Ollama: Download & Install Ollama
- Run
ollama pull phi3to download the default model.
- Run
- Clone the repository:
git clone https://github.com/YASH-810/ARIA.git cd ARIA - Create and activate a virtual environment:
python -m venv venv .\venv\Scripts\activate
- Install dependencies:
pip install -r requirements.txt
Simply run the bootstrapper:
.\start.batThe first run will automatically download the Piper TTS engine and voice models (~50 MB).
Press F2 at any time to trigger voice listening. ARIA will listen until you stop speaking, then transcribe and respond instantly.
| Command | Description |
|---|---|
/mute |
Disable voice output |
/unmute |
Enable voice output |
/model <name> |
Switch the active Ollama model |
/context on/off |
Toggle persistent conversation memory |
/debug on/off |
Toggle verbose logging |
/state |
View current system state |
/help |
Show all available commands |
ARIA can execute specific intents instantly without calling the LLM:
open notepad→ Launches Windows Notepadsearch for weather in London→ Opens browser to searchplay lofi on youtube→ Launches YouTube searchrun dir→ Executes shell command
graph TD
User([User Input]) --> CLI[CLI - prompt_toolkit]
CLI --> Orchestrator{Orchestrator}
Orchestrator -- Fast Path --> Router[Router / Fuzzy Matcher]
Router --> Tools[System Tools / Apps]
Orchestrator -- Slow Path --> Engine[LLM Engine - Ollama]
Engine --> Pipeline[Voice Pipeline]
Pipeline --> STT[STT - faster-whisper]
Pipeline --> TTS[TTS - Piper]
TTS --> Sync[Terminal Sync Output]
Sync --> User
- Vision Support: Screen interaction via OCR & CV.
- Wake Word: Always-on listening for "Hey ARIA".
- Dashboard: Electron-based React UI for visual status tracking.
- Custom Tools: Easy-to-use API for adding third-party integrations.
ARIA is built on the principle of Local First. All audio processing, transcriptions, and LLM inferences happen on your local machine. No voice data or chat history is ever uploaded to external servers.