OpenFlip pocket agent
- Ashikka unveiled OpenFlip, a pocket AI hacking agent that scans RF, Bluetooth, Wi‑Fi, and NFC and writes exploit modules on the fly. - The device won first place at an OpenAI Codex Hackathon and opened a waitlist after a demo video. - It targets hardware and security side projects by combining on‑device sensing with agentic code generation for modular exploits (x.com).
A radio scanner is a sensor for invisible signals, and Ashikka Gupta’s OpenFlip packs several of them into a handheld device that can also generate attack code for what it finds. (github.com) (x.com) The public GitHub repository describes OpenFlip as a “personal pocket red teaming agent,” and the project files frame it as a Flipper Zero-style device with an open registry for downloadable modules. (github.com 1) (github.com 2) In the project’s design brief, OpenFlip is pitched as “a pocket-sized AI red-team agent” that “dynamically installs exploit modules from an open registry,” with an 18-second demo loop showing signal detection, registry search, module install, verification, and decoding. (github.com) Red teaming is security testing that imitates an attacker, and devices in this category usually start by listening for nearby protocols such as radio frequency, Bluetooth, Wi‑Fi, or near-field communication. OpenFlip’s pitch combines that hardware scanning step with software that writes or pulls the code needed to probe a target. (github.com 1) (github.com 2) That setup puts OpenFlip in the same general family as Flipper Zero, a handheld tool known for interacting with access-control and wireless systems, but Gupta’s materials describe an added layer: an autonomous agent that can learn new capabilities from the internet on demand. (github.com) The timing lines up with a broader push toward agentic coding tools. OpenAI says Codex is built for “real engineering work,” and its Codex app is designed to manage multiple coding agents across projects and long-running tasks. (openai.com 1) (openai.com 2) Cerebral Valley’s public gallery for the OpenAI Codex Hackathon shows 53 submissions and several finalists, confirming the event where OpenFlip was presented, though the gallery page visible in search does not list OpenFlip in the excerpted winner section. (cerebralvalley.ai) Gupta’s repository also includes a waitlist landing-page prompt for OpenFlip.io, with copy that says the device “listens, learns new tricks from the internet, and speaks back.” That language captures the project’s core bet: treat hardware hacking less like loading fixed tools and more like calling in software that writes the next tool when needed. (github.com)
