Ways to Talk & Interact
Ways to Talk & Interact
Project AIRI is designed to feel alive across multiple sensory channels. You are not limited to typing in a standard chat box—you can speak aloud via push-to-talk, engage in continuous natural conversations with open-mic voice activity detection, experience sub-second real-time streaming with Gemini Live, or chat through Discord.
1. Interaction Modes Overview
| Interaction Mode | Input Method | Audio Latency | Ideal Use Case |
|---|---|---|---|
| Desktop Chatbox | Keyboard / Text input | N/A (Instant text) | Detailed roleplay, tweaking prompts, issuing complex multi-step commands. |
| Push-to-Talk (PTT) | Hold Hotkey + Microphone | ~1–2s (STT → LLM → TTS) | Precise voice commands while gaming or working without accidental mic pickup. |
| Open Mic (VAD) | Voice Activity Detection | ~1–2s | Hands-free conversations while your desktop companion sits on screen. |
| Gemini Live Bidi | Real-Time WebSocket Audio | < 500ms (Sub-second) | Instant fluid banter with natural human-like voice inflections and mid-sentence interruptions. |
| Discord Voice & Chat | Slash Commands & Voice Channel | ~1–3s | Hanging out with friends and your AI companion in a shared Discord server. |
2. Desktop Chatbox
The persistent Desktop Chatbox is your primary hub for deep interactions.
- Workspace Layout: The chat surface features a 3-column workspace layout:
- Left Sidebar: Navigation drawer with shortcuts to the Studio, Media Library, Eternal Memory Thread, and Event Ledger.
- Main Conversation Stream: Interleaved message bubbles, tool-call cards, and emotion tokens.
- Right Context Rail: Shows active memory chips (Echo Chips), current mood, and active character cards.
- Journal Chips: Text and image journal entries written by AIRI appear inline as interactive chips. Clicking a chip opens the full entry or image preview.
- Composer Toolbar: Use the bottom composer bar to send messages, attach images for vision analysis, trigger suggestion prompts (AI Producer), or toggle between characters.
3. Voice Input: Push-to-Talk vs. Open Mic
You can talk to AIRI directly using your microphone.
Setting Up Hearing (STT)
- Open Settings → Modules → Speech-to-Text (Hearing).
- Select your STT provider (e.g., Local Whisper WebGPU, Groq Whisper, OpenAI Whisper, or Cloudflare Workers AI).
- Select your preferred microphone in the Audio Devices selector.
Push-to-Talk (PTT)
- Press and hold your configured PTT key (configured in Settings → General → Shortcuts).
- Speak your message. When you release the key, AIRI transcribes your speech, sends it to the mind model, and replies with synthesized voice.
Open Mic with Voice Activity Detection (VAD)
- When enabled in the Control Strip or Settings, AIRI listens continuously.
- The built-in Silero VAD / WebRTC VAD detects when you start speaking and automatically finalizes the turn when you pause.
TIP
VAD Sensitivity: If AIRI cuts you off too early, increase the Silence Detection Threshold in Hearing Settings to allow for natural pauses while thinking.
4. Gemini Live API (Bidirectional Audio)
For ultra-low latency, conversational audio that feels like a real phone call, AIRI supports Google's Gemini Live Bidi WebSocket streaming.
What Makes Gemini Live Different?
- Native Audio-to-Audio: Instead of separate STT → Text LLM → TTS hops, Gemini processes your raw audio PCM stream and generates conversational speech audio directly.
- Sub-Second Latency: Audio starts streaming back in under 500ms.
- Natural Interruptions: If AIRI is speaking and you start talking, the audio output immediately stops, yielding the floor back to you naturally.
How to Enable Gemini Live
- Go to Settings → Providers → Google Gemini and enter your Gemini API Key.
- Open the Gemini Live modal from the Control Strip or Chatbox header.
- Choose a voice model and click Connect.
- Speak freely without needing push-to-talk triggers.
5. Discord Integration
AIRI can run as a Discord bot on your personal server:
- Text Chat: Mention
@AIRIor reply to her messages in any authorized channel. - Slash Commands: Use
/chat,/steer,/voice join, or/voice leaveto control the bot. - Voice Channels: When in a voice channel, AIRI listens to users speaking and responds with synthesized voice directly into the Discord voice room.
(See the Discord Commands Reference for full setup instructions.)
6. Interruptions & Stopping Output
If AIRI is speaking or generating a long response and you want to stop her:
- Click the Stop button on the floating Control Strip or Chat composer.
- In Gemini Live mode, simply speak aloud—the system detects your voice and cancels in-flight playback immediately.