Image: dev.to
Cracking a robot vacuum's cloud API and building a dashboard
A developer reverse engineered a robot vacuum's cloud request-signing and Dart-compiled app using traffic capture, decompilation and Frida hooks. Claude helped read dumps, write hook scripts and keep notes across sessions. The author built a dashboard that goes beyond the official app and used Claude vision to filter drive commands.
The full story
Federico Sciuca's 3i G10+ robot vacuum (about €400) has lidar, a camera, a live room map and 32 different cloud commands, but the official app shows its owner very little of it. He started out wanting just the battery level on his own dashboard. A week of late nights later, he had worked out the cloud's request-signing algorithm, matching 18 out of 18 captured requests exactly.
Using Frida hooks in a patched copy of the Flutter app, he captured the live map data and the encrypted video-stream credentials. He then built a 5,000-line Python dashboard with video streaming, cost tracking, behaviour rules and an autonomy loop. Claude helped read the dumps, write the hook scripts and keep notes between sessions. He eventually gave Claude motor control, with Claude Haiku's vision checking the camera feed (about $0.003 per call) so the robot stopped crashing. Peak AI spend was $0.89 a day.
Author-reported; the code is private. The author also describes working around vendor access controls and says some of the approach is "not elegant". Exposed map and camera-related data is a privacy issue for owners.
How it was done
Tools
Themes
Loading themes…
Sources
From the maker's own write-up or code. We summarise what they report; we haven't tested it ourselves.
- DEV: I cracked a robot vacuum's API in a week and gave Claude the keys
Detailed write-up: signing algorithm, Frida hooks, the dashboard and handing driving to Claude.