// the find
evilsocket/pwnagotchi
(⌐■_■) - Deep Reinforcement Learning instrumenting bettercap for WiFi pwning.
Pwnagotchi is a Raspberry Pi-based tool that uses A2C reinforcement learning to autonomously capture WPA handshakes via bettercap. It's a WiFi security auditing gadget with an e-ink display that shows a tamagotchi-style face while it works. Aimed at penetration testers and people who want to run passive WiFi audits over time while commuting or walking around.
The RL approach is genuinely interesting — the agent tunes deauth timing and channel hopping based on observed handshake yield, which is more adaptive than fixed scripts. Multi-unit mesh cooperation (units advertise presence via custom 802.11 IEs and split channels between them) is a clever idea that costs nothing extra if you already have multiple devices. The plugin system is clean and the community has built a long tail of plugins for GPS logging, WiGLE upload, Bluetooth tethering, and more. Hardware support is unusually broad — a dozen Waveshare and DFRobot e-ink panels are all first-class, not afterthoughts.
The repo is built on stable-baselines, which was deprecated in favor of stable-baselines3 years ago; the underlying RL code hasn't kept pace with the ecosystem and the 'AI' is closer to parameter tuning than anything modern practitioners would call deep RL. Setup complexity is high — you're flashing a custom Pi image, configuring monitor mode, and managing bettercap as a subprocess, and the docs assume you already know what you're doing. The last meaningful architectural work was 2020-era; recent commits are mostly locale files and hardware driver additions, not core improvements. Committed `.pyc` files in the tree are sloppy and suggest the repo isn't under tight review.