nethical6/conversation-steganography — explained in plain English
Analysis updated 2026-05-18
Send a hidden message disguised as normal chat text through any messaging app.
Experiment with AI-based steganography for research or educational purposes.
Test the encode and decode process locally using the built in simulation mode.
Study how a local language model can encode encrypted bytes into its word choices.
| nethical6/conversation-steganography | cosmtrek/mindwalk | oracle/terraform-provider-oci | |
|---|---|---|---|
| Stars | 940 | 889 | 865 |
| Language | Go | Go | Go |
| Last pushed | — | — | 2026-07-15 |
| Maintenance | — | — | Active |
| Setup difficulty | moderate | easy | moderate |
| Complexity | 4/5 | 2/5 | 4/5 |
| Audience | researcher | developer | ops devops |
Figures from each repo's GitHub metadata at analysis time.
Requires building from Go source and downloading a local AI model during first-run setup.
Conversation Stenography is a command line tool that lets two people hide a private message inside text that looks like an ordinary chat conversation. You type a secret message, and the tool uses a local AI language model to generate a completely different, innocent sounding sentence that gets copied and pasted into a normal messaging app like WhatsApp, Telegram, Signal, iMessage, or email. On the other end, the recipient's copy of the same tool decodes that innocent looking text back into your original secret message. The project explains that it exists because of concern over governments scanning private messages, and because sending obviously encrypted text can draw suspicion, while this approach hides the fact that a secret message was sent at all. The author is upfront that this is a proof of concept built by an eighteen year old, that similar techniques using language models have existed for years, and that the project has known weaknesses since researchers are actively developing ways to detect hidden content inside AI generated text. The README labels the project educational and research use only, and states the creator takes no responsibility for misuse. Under the hood, your message is first encrypted using AES-SIV, described as authenticated encryption, and every message is cryptographically linked to the one before it so that tampering, deletion, or reordering can be detected. A local AI model, which runs entirely on your own device rather than in the cloud, then generates the cover text by encoding the encrypted bytes into its word choices, and the same model on the recipient's device reverses that process. Both people need to meet in person beforehand to agree on a shared secret phrase, which is turned into an encryption key and never saved to disk. To use it, you build the tool from source using Go, run a setup wizard that helps you choose and automatically download an AI model suited to your computer, and then start chatting through a terminal interface where typing a message generates cover text to copy elsewhere, and a paste command decodes messages you receive. A built in simulation mode lets you test the whole process with two simulated users on a single device before trying it with a real friend.
A command line tool that hides encrypted secret messages inside AI-generated, innocent-looking chat text you paste into any messaging app.
Mainly Go. The stack also includes Go, AES-SIV, PBKDF2.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
double-check against the repo, no cap.