🔉 Introducing Ultrasonic Agentics: Hide Secret AI Commands & Data in Plain Sound & Video
This weekend I built a completely hidden Ai Commands and Control Data communication system embedded in the background noise, completely invisible to detection. I created this using Claude Code and my SPARC automation system. Total spend? Just the $200/month Claude Max subscription. No infrastructure, no compute clusters, just recursive loops and a stream of thought turned into code.
Ultrasonic Agentics is a steganographic framework that embeds encrypted AI commands into ultrasonic frequencies (18–20 kHz), completely inaudible to humans but crystal clear to software. You can transmit secure agent commands through Spotify, YouTube, AM/FM radio, or even intercoms, fully encrypted, undetectable, and energy-efficient.
I used Claude-SPARC to iterate the entire system: command-line tools, REST API, real-time stream encoder, web UI, and even an MCP server. AES-256-GCM handles encryption, FSK modulation takes care of the signal, and decoding runs with less than 1KB of RAM. It supports batch or real-time processing, works on ESP32s, and includes psychoacoustic masking for seamless media integration.
Perfect for smart home control, covert agent comms, emergency fallback channels, or just embedding secret triggers in everyday media.
What started as a fun experiment is now a fully working protocol. Secure, invisible, and already agent-ready.
Install from PyPI
pip install ultrasonic-agentics
What is it?
Ultrasonic Agentics is a cutting-edge steganography framework that embeds encrypted AI commands into ultrasonic frequencies (18-20 kHz) - inaudible to humans but detectable by your applications. Perfect for secure command transmission, covert communication channels, and innovative AI agent coordination.
100% untraceable ultra-secret communications at super low power. Embed AI commands into any sound or video file and broadcast through any medium - unlicensed spectrum, VHF, AM/FM radio, Spotify streams, YouTube videos, or any audio channel. Commands hide in the background noise, completely invisible to detection.
✨ Key Features
🎯 Use Cases
Download and Try
🏗️ Architecture
Ultrasonic Agentics uses a sophisticated signal processing pipeline:
🔋 Low-Power Command & Control
Ultrasonic Agentics is designed for energy-efficient operation, making it ideal for battery-powered and embedded systems:
Power Advantages
Ideal for Embedded Systems
🔒 Security Features
🛠️ Quick Start
Installation
# Install from PyPI
pip install ultrasonic-agentics
# With all features
pip install ultrasonic-agentics[all]
Command Line Interface
After installation, three CLI tools are available:
ultrasonic-agentics - Main CLI
# Show help and available commands
ultrasonic-agentics --help
# Embed a command in an audio file
ultrasonic-agentics embed -i input.mp3 -o output.mp3 -c "command:execute" -k your-secret-key
# Embed with custom frequency and amplitude
ultrasonic-agentics embed -i input.mp3 -o output.mp3 -c "deploy:v2" \
--freq 19000 --amplitude 0.05 --bit-duration 0.02
# Decode commands from audio
ultrasonic-agentics decode -i output.mp3 -k your-secret-key
# Decode with verbose output
ultrasonic-agentics decode -i output.mp3 -k your-secret-key --verbose
# Analyze audio for ultrasonic content
ultrasonic-agentics analyze -i audio.mp3
# Analyze with spectrogram output
ultrasonic-agentics analyze -i audio.mp3 --spectrogram --output report.png
# Configure default settings
ultrasonic-agentics config --freq 19000 --bit-rate 500
# Show current configuration
ultrasonic-agentics config --show
ultrasonic-server - MCP Server
# Start MCP server for AI agent integration
ultrasonic-server
# With custom port
ultrasonic-server --port 8080
ultrasonic-api - REST API Server
# Start REST API server
ultrasonic-api
# With custom configuration
ultrasonic-api --host 0.0.0.0 --port 8000 --workers 4
CLI Quick Reference
CommandDescriptionExampleembedHide command in audioultrasonic-agentics embed -i in.mp3 -o out.mp3 -c "cmd"decodeExtract hidden commandultrasonic-agentics decode -i out.mp3 -k keyanalyzeDetect ultrasonic contentultrasonic-agentics analyze -i audio.mp3configManage settingsultrasonic-agentics config --show
Common Options:
Basic Usage
from agentic_commands_stego import AudioEmbedder, AudioDecoder
# Embed a command
embedder = AudioEmbedder()
command = "execute: deploy_model --version 2.0"
secure_audio = embedder.embed_from_file("original.mp3", command)
secure_audio.export("output.mp3", format="mp3")
# Decode the command
decoder = AudioDecoder(embedder.cipher.key)
decoded_command = decoder.decode_from_file("output.mp3")
print(f"Hidden command: {decoded_command}")
🎮 Interactive Web Interface
Experience Ultrasonic Agentics through our modern React-based UI:
# Start the web interface
cd ui && npm install && npm run dev
Visit http://localhost:5173 to:
🤖 MCP Integration
Use Ultrasonic Agentics with AI agents via Model Context Protocol:
# Start the MCP server
ultrasonic-server
# Use with Claude or other MCP-compatible AI
ultrasonic-agentics encode "AI: process customer data" audio.mp3
ultrasonic-agentics decode audio.mp3
🔧 Advanced Features
Streaming API
# Real-time encoding for live audio
from agentic_commands_stego import StreamEncoder
encoder = StreamEncoder()
for chunk in audio_stream:
encoded_chunk = encoder.process(chunk, command_queue.get())
output_stream.write(encoded_chunk)
REST API
# Start the API server
ultrasonic-api
# Embed via API
curl -X POST http://localhost:8000/embed \
-F "audio=@input.mp3" \
-F "command=deploy:production" \
-F "key=your-secret-key"
Video Support
# Embed in video files
from agentic_commands_stego import VideoEmbedder
embedder = VideoEmbedder()
embedder.embed_from_file(
"video.mp4",
"AI: analyze frames for objects",
"output.mp4"
)
📝 Examples
Quick Start with Sample Files
The /examples directory contains sample media files for testing:
# Navigate to examples directory
cd examples/
# Test embedding a command in the sample audio file
ultrasonic-agentics embed -i sample_audio.mp3 -o audio_with_command.mp3 -c "hello world"
# Decode the embedded command
ultrasonic-agentics decode -i audio_with_command.mp3
# Test with the sample video file
ultrasonic-agentics embed -i sample_video.mp4 -o video_with_command.mp4 -c "AI: process video"
ultrasonic-agentics decode -i video_with_command.mp4
Python Examples
Basic Encoding/Decoding
from agentic_commands_stego.embed.audio_embedder import AudioEmbedder
from agentic_commands_stego.decode.audio_decoder import AudioDecoder
# Create embedder with a specific key
key = b'my-secret-key-32bytes-padding!!!' # 32 bytes for AES-256
embedder = AudioEmbedder(key=key)
decoder = AudioDecoder(key=key)
# Embed a command
success = embedder.embed_file(
'examples/sample_audio.mp3',
'output_with_command.mp3',
'execute: deploy --version 2.0'
)
# Decode the command
command = decoder.decode_file('output_with_command.mp3')
print(f"Decoded: {command}")
Real-time Audio Processing
from agentic_commands_stego.decode.audio_decoder import AudioDecoder
# Set up real-time listener
decoder = AudioDecoder(key=your_key)
def on_command_detected(command):
print(f"Detected command: {command}")
# Process the command
if command.startswith("execute:"):
action = command.split(":", 1)[1]
# Perform action
# Start listening through microphone
decoder.start_listening(callback=on_command_detected)
Batch Processing
import os
from agentic_commands_stego.embed.audio_embedder import AudioEmbedder
embedder = AudioEmbedder()
# Process multiple files
audio_files = ['file1.mp3', 'file2.wav', 'file3.ogg']
commands = ['cmd1', 'cmd2', 'cmd3']
for audio_file, command in zip(audio_files, commands):
output = f"secured_{audio_file}"
embedder.embed_file(audio_file, output, command)
print(f"Processed: {audio_file} -> {output}")
Custom Frequency Configuration
from agentic_commands_stego.embed.ultrasonic_encoder import UltrasonicEncoder
from agentic_commands_stego.decode.ultrasonic_decoder import UltrasonicDecoder
# Use lower frequencies (17-18 kHz) for better speaker compatibility
encoder = UltrasonicEncoder(
freq_0=17000, # Frequency for bit '0'
freq_1=18000, # Frequency for bit '1'
amplitude=0.15, # Slightly higher amplitude
bit_duration=0.02 # Slower bit rate for reliability
)
decoder = UltrasonicDecoder(
freq_0=17000,
freq_1=18000,
detection_threshold=0.05
)
# Encode and decode
signal = encoder.encode_payload(b"low-freq test")
decoded = decoder.decode_payload(signal)
Steganography for AI commands, cool. But what about accidental decoding by non-target devices?
Imagining smarter healthcare; building bridges; burning siloes
2moMost of those transmission modes are designed for human hearing so they aggressively filter those high frequencies you are using. No doubt error rates would be high for those.
Head of Application Software at DATA MODUL AG
2moSounds good, but since the majority of all transmitted audio IS lossy, I doubt someone can remote control your home-devices. You already point it out in your repo: does not work well with MP3/OGG/AAC Reuven Cohen
Reuven Cohen you might get a call: "we want you to meet us at this restaurant, take a pair of socks and good mood - we'll give you a mission!" 🤣😂