⬇️ Desktop Client

Avatar Desktop Client v2

The lightweight Windows app that feeds your screen to your AI avatar companion — so it can react live on stream.

🖥️
StreamTracker Avatar Client v2
Windows 10 / 11 ~15 MB Python 3.11+
⬇️ Download for Windows (.exe) 🍎 Download for macOS (.zip)
Windows: no installation — run the .exe directly. SmartScreen may warn on first run; click "More info → Run anyway".
macOS: needs Python 3 (one-time install), then:
  1. Unzip the download
  2. Right-click RUN_MAC.command → Open → Open (first time only; after that just double-click)
  3. Say yes to the Screen Recording, Microphone, and Input Monitoring permission popups (Input Monitoring is what makes the talk hotkey work while you're in-game — if you missed it: System Settings → Privacy & Security → Input Monitoring → enable your terminal, then relaunch)
Full walkthrough is in the README_MAC.md inside the zip. Voice push-to-talk: hold fn+F5, use the Hold-to-talk button on the Advanced tab, or the browser panel.
🤖 Bringing your own AI agent? Grab the co-host skill — paste it into your agent (OpenClaw skill, system prompt, anywhere) and it knows the whole session loop: instant voice reactions, when to look at your screen, credit etiquette. Want your agent to stay itself instead of playing a character? Use the minimal loop skill — just the react-to-voice loop, nothing else. MCP-native agents can also pull either directly from the server (prompt names: cohost, loop).
What's new
v2.3.1  — macOS + bring-your-own-agent
• macOS support: native screen grab (no OBS required), one-command setup via RUN_MAC.command, and a global push-to-talk hotkey built on Apple's Quartz event taps (grant Input Monitoring once). Extensive stability work: connect freeze, emoji rendering crash, and voice-enable crash all fixed.
• Agent mode: check one box and your own AI (via the MCP server) does the thinking — the built-in brain stays silent. Frame feed selectable down to 6 seconds (free, no credits).
• Every push-to-talk line now ships with a screenshot of that exact moment, skipped automatically when the screen hasn't changed.
• Your agent can demand a live screenshot any time with get_latest_frame {"fresh": true}.
• Hold-to-talk button + window hotkey as no-permission-needed voice fallbacks; fixed mic stutter while holding the talk key.
• Older API keys (pre-st_) and pasted panel URLs now accepted at Connect.
v2.2.0  — UI overhaul
• Sidebar navigation replaces the old top tabs — clearer hierarchy with icons + active accent stripe.
• Persistent quick-actions bar: Start/Stop capture and Capture now from any tab.
• Collapsible activity log (toggle with Ctrl + /) frees up vertical space.
• Tighter, more modern spacing and typography across all panels.
• Builds are now produced automatically by GitHub Actions for every release.
v2.1.0  — Voice + chat
• Push-to-talk voice with ElevenLabs Scribe transcription bundled into the next capture.
• Twitch chat command + raid/sub/cheer event triggers with cooldown.
• Game-state agents (LoL, Steam, Xbox) feed structured context into prompts.
v2.0.0  — Initial v2 release
• Rewritten desktop client with dark theme, OBS WebSocket capture, scene-change skipping, and panel bridge for the browser overlay.
What's in v2
📸
Auto Screen Capture
Captures your desktop on a configurable interval (1 – 15 min) and sends it to your vision AI.
🎙️
Push-to-Talk Voice
Hold a hotkey to speak directly to your avatar. The transcript is included with the next capture.
🎬
OBS Source Capture
Connect via OBS WebSocket to capture a specific scene source instead of the whole desktop.
⏱️
Live Countdown Timer
See exactly when the next capture fires. Skip unchanged scenes automatically to save API credits.
🎮
Game State Context
Paste custom context — current boss, score, etc. — so the avatar's comments are more relevant.
🔗
Panel Bridge
Connects to the browser overlay panel over WebSocket for real-time avatar animations.
Getting Started
1

Download & Run

Download StreamTracker-Avatar-v2.exe above and double-click it. No installation needed.

Windows SmartScreen may flag unsigned executables. Click More info → Run anyway to proceed. The app makes no system changes.
2

Paste Your API Key

Open the Connect tab and paste your StreamTracker API key (starts with st_). Click Connect — your avatar configuration loads automatically.

Get your API key from Account → API Keys on streamtracker.cloud.
3

Configure Capture

On the Capture tab, choose your capture mode (desktop or OBS source), set an interval, and optionally enable voice capture. Click Start Session.

4

Watch the Session Tab

Switch to Session to see your avatar's live comments as they come in. The feed updates in real time — no page refresh needed.

System Requirements
Component Requirement
OS Windows 10 or Windows 11 (64-bit) Required
Internet Active internet connection to streamtracker.cloud Required
API Key StreamTracker account with an active API key Required
OBS WebSocket OBS Studio 28+ (for OBS source capture mode) Optional
Microphone Any input device (for push-to-talk voice) Optional
Running from Source (Mac / Linux)
Python 3.11+ required. Clone the repo, then:

cd desktop_client_v2
pip install -r requirements.txt
python main.py

The GUI uses Tkinter (included with most Python distributions). On Ubuntu/Debian: apt install python3-tk

All set up? Head to the Studio

Configure your avatar's personality, vision provider, voice, and overlay style in Avatar Studio.

🎭 Open Avatar Studio 📖 Setup Guide ← Dashboard