# OBS Agentic Control Interface

> Source: <https://github.com/DeepBlueDynamics/obsagent>
> Published: 2026-06-17 00:08:41+00:00

Note

**🤖 100% Coded by AI**: This entire repository and application was engineered 100% autonomously by **Antigravity**, an agentic AI coding assistant.

A powerful, self-contained agentic interface built in Rust that allows you to control OBS Studio via natural language. The system features an advanced reasoning loop powered by Claude (`claude-3-5-sonnet`

) and OpenAI (`gpt-4o`

), automatically connected to OBS Studio via WebSocket v5.

**📊 Real-Time Interactive Dashboard**: Monitors streaming, recording, virtual camera, active program scenes, and audio mixer volume levels dynamically.**🖥️ Canvas Size & Output Resolution Detection**: Automatically queries your OBS Canvas and scaled output resolutions, displaying them in a dedicated status card and feeding them into the AI agent's system prompt context.**🧠 Intelligent Auto-Fallback Agent Loop**:- Routes requests through a robust agent loop with support for both Claude and OpenAI.
**Dynamic failover**: If the selected provider key or network request fails, the backend automatically switches to the alternative provider (Claude ⇄ OpenAI) and finishes your request seamlessly without duplicating message history.

**🔌 Windows Hotwire & WGC Binding**:- Remotely scans, restores, focuses, and resizes target application windows on your host OS.
- Automatically binds windows into OBS using
**Windows Graphics Capture (WGC / Windows 10 method)** to prevent black/grayed-out capture screens. - Uses an
**intelligent fuzzy resolver** to match window titles/classes against OBS's active window pool to ensure 100% reliable binding.

**🎙️ Real-Time Voice Activation ("OBSy")**:- Uses browser-side VAD (Voice Activity Detection) and OpenAI's Whisper API for high-fidelity speech-to-text.
- Triggers on the wake word
`"OBSy"`

. Allows natural language control directly from your microphone.

**🎛️ Manual Sidebar Controls**: Instant buttons to trigger scene changes, volume adjustments, and transitions that stay fully in sync with the AI's state.

**OBS Studio**(v28.0 or later, featuring native WebSocket support).** Rust Toolchain**(pre-configured in this workspace environment).** GitHub CLI (**(for repository authentication and deployment).`gh`

)

- Open OBS Studio.
- Navigate to
**Tools**➔** WebSocket Server Settings**. - Check
**Enable WebSocket server**. - Take note of the port (default:
`4455`

) and the password (or generate a new one).

To start the server, configure your credentials and connection endpoints as environment variables:

```
# Set your API keys (one or both are supported)
export ANTHROPIC_API_KEY="your-anthropic-api-key"
export OPENAI_API_KEY="your-openai-api-key"
export ELEVENLABS_API_KEY="your-elevenlabs-api-key"

# Set the connection host for OBS Studio
# When running inside a container, set to "host.docker.internal" to connect to the host
export OBS_HOST="host.docker.internal"
export OBS_PORT="4455"
export OBS_PASSWORD="your-obs-websocket-password"
```

To compile and launch the Axum web server, run the following in the workspace directory:

```
cargo run
```

The server will initialize and serve the frontend dashboard at `http://localhost:8080`

.

Open your browser, connect OBS, and start commanding your streams via text or voice!