De-Clouding IoT Devices: Local Control for Midea Air Conditioners and Tuya-Based Cat Feeders
A detailed technical guide demonstrates how to remove cloud dependency from consumer IoT devices by reverse engineering their hardware and replacing vendor firmware.
The author had long preferred Zigbee devices for a smart home to avoid vendor clouds but was compelled to purchase Wi-Fi-only appliances after adopting a cat. These included a Midea air conditioner paired with a Daichi controller and a Smart Life cat feeder, both built on Tuya platforms. The primary concern was lack of control over devices containing microphones and other sensors that phone home.
Hardware Toolkit and Initial Analysis
Essential tools included a multimeter, soldering iron, UART adapter supporting 3.3V/5V, flux, solder wick, and optional hot air station. The researcher emphasized practicing soldering on scrap boards first.
The Daichi DW23-B controller for the air conditioner contained a TYWE3S module based on ESP8266. The onboard USB port functioned as a UART interface, controlled by a physical switch. Firmware was backed up using esptool.py and replaced with ESPHome, enabling native climate control in Home Assistant.
Cat Feeder Reverse Engineering
The feeder used a WBR3 module on an RTL8720CF chipset communicating with an MCU via UART in the classic TuyaMCU architecture. After identifying UART pins with a multimeter and capturing traffic using TuyaMCUAnalyzer, the Wi-Fi controller was desoldered to access programming pins A_0 and A_15.
Firmware was read and written with ltchiptool. The device was then flashed with OpenBeken, which supports TuyaMCU protocol emulation. Configuration via autoexec.bat enabled drivers for TuyaMCU and NTP, mapped dpIDs to channels for battery level, feeding button, and scheduling, and exposed the device over MQTT for Home Assistant integration.
A custom automation script was created to simulate button presses for manual feeding, while schedule data structures were analyzed from raw protocol packets for potential future local implementation.
Related articles
Keurig Coffee Machine Scans Home Network and Consumes 1TB of Data in Ten Days
A user reported that a Keurig smart coffee machine belonging to his father consumed approximately 1TB of network traffic over ten days, primarily by scanning the local home network. The incident highlights how everyday IoT devices such as coffee makers, refrigerators, and air fryers are now collecting user data for resale to advertising networks. The technology traces its roots to Automatic Content Recognition (ACR) first popularized by Shazam and later adopted by television manufacturers including Samsung, LG, Vizio, and TCL. These devices capture screen fingerprints every second and transmit them to remote servers, building detailed viewing profiles that are sold to third parties. Similar data collection practices have expanded to kitchen appliances through platforms like KitchenOS, which aim to create lifelong customer relationships by monetizing attention and behavioral data. Most users unknowingly consent to this surveillance because the relevant options are buried in lengthy terms of service and pre-selected during setup. Experts note that the volume of traffic observed on the Keurig device suggests either aggressive network scanning or a possible software bug in the data collection module.
Why Split-Tunneling Leaks Persist in Android VPN Clients and How Work Profiles Provide a Fix
Popular Android VPN clients such as Happ, v2rayNG and Hiddify cannot reliably combine split-tunneling with a true kill-switch because of AOSP limitations in VpnService.Builder and netd. When the system Lockdown mode is enabled, applications excluded from the tunnel lose all connectivity; when it is disabled, any client crash, LMKD kill, network switch or boot delay allows traffic from apps like Claude to bypass the VPN and expose the real IP. Alternative approaches using full-tunnel plus Xray routing or AdGuard DNS plus SNI proxy also fail on Android due to QUIC, DoH, hidden telemetry domains and ConnectivityManager flags that break banking apps. Anthropic’s strict regional enforcement means a single leaked request can permanently ban a $200 Claude Max subscription with no refund. The only architecture that achieves fail-closed isolation places Claude and ChatGPT inside a dedicated Work Profile managed by Shelter and Pocket Android, where Profile Owner rights allow Always-on VPN plus Lockdown to be applied exclusively to user 10 while the main profile remains untouched.
Purism Announces Librem 5 v2 Smartphone on RISC-V with Hardware Kill Switches for Camera, Microphone and Connectivity
Purism has revealed the second-generation Librem 5 smartphone, marking a major shift from Arm to RISC-V architecture with the eight-core SpacemiT K1 processor. The device retains its core privacy focus through physical hardware switches that can disable the camera, microphone, GPS, Wi-Fi, Bluetooth and cellular modem. Both the Wi-Fi/Bluetooth module and the modem are designed as removable cards, allowing users to replace or completely remove them, with future plans for 5G and NFC modules. Specifications include a 5.7-inch 1440x720 display, 13 MP and 8 MP cameras, a 4500 mAh removable battery, 8 GB RAM and 128 GB storage, dual SIM support, microSD slot and headphone jack. The phone runs PureOS with the Phosh shell based on GNOME and supports desktop mode when connected to an external monitor. Purism will not add fingerprint or facial recognition due to legal concerns in the United States, pricing the device at $1,599 with developer kits expected after a crowdfunding campaign.
Four VLESS VPN Configurations Show Varying Lifespans Before Network Restrictions
A Habr author under the nickname PXI published results from a private experiment involving four VPN stands running the VLESS protocol. The test compared REALITY and Self-SNI TLS setups, both directly and through an intermediate server located in Russia. One REALITY configuration with an intermediate server operated for four months until manually disabled by the author. Direct REALITY connections encountered restrictions after approximately three months. The two Self-SNI TLS stands became unavailable after eight days and three days respectively. In some cases IP accessibility later recovered, suggesting possible subnet-wide blocking. The author attributes detection to metadata analysis such as packet sizes, inter-packet intervals, connection duration and traffic direction ratios rather than confirmed active probing or machine learning by DPI systems.