This project enables private conversations by hiding secret messages within seemingly normal chat text transmitted via any messaging app. It targets users concerned about surveillance, offering a practical LLM-based steganography solution that encrypts messages locally and disguises them as innocent cover text, undetectable to intermediaries.

How It Works

Secret messages are first AES-SIV encrypted. A local AI model then generates natural-sounding cover text, embedding the encrypted bytes within its token choices. This cover text is sent through standard messaging platforms. The recipient's instance uses the same AI model to extract the hidden bytes, which are then decrypted back to the original message. This ensures end-to-end privacy without cloud reliance.

Quick Start & Requirements

Installation involves cloning the repo and building the Go binary: git clone https://github.com/nethical6/conversation-steganography.git && cd conversation-stenography && go build -o conversation-stenography ./cmd/conversation-stenography. Prerequisites include Go 1.22+ and Python 3.9+. The setup wizard (./conversation-stenography) guides model selection/download and configuration. A local simulation mode (./conversation-stenography simulate) allows testing.

Highlighted Details

• Employs AES-SIV for authenticated encryption, ensuring confidentiality and integrity.
• Utilizes local AI models (e.g., Llama 3.2, GPT-2) running entirely on-device, preventing cloud data leakage.
• Implements a cryptographically linked conversation chain to detect message tampering.
• Supports multiple AI model runtimes, including MLX for Apple Silicon and Transformers.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), or active development signals were found in the provided README.

Licensing & Compatibility

The project is provided for "Educational Use Only" and as a "proof of concept." The specific license in the LICENSE file may impose restrictions on commercial use or integration into closed-source applications.

Limitations & Caveats

This project is explicitly for educational/research purposes, with a disclaimer against unauthorized activities. It's a proof of concept, and advanced techniques may detect hidden content. Both users must use identical AI models and process messages strictly in order for correct decoding.