Own Xiaomi scale, one agent prompt: BrOrlandi’s local BLE dashboard
Bruno Orlandi (@BrOrlandi) posted a public agent reverse-engineering story: a local dashboard for the Xiaomi scale he already owns, built from one coding-agent prompt, then published as MIT source. rea.run credits him and the open-source dashboard.
Short version
He wanted the scale’s readings on his own server — training context from a watch, meal ideas — without staying inside the vendor app. The follow-up demo says one prompt produced an app he preferred to the official one. The repo is a self-hosted listener: while someone stands on his scale, a nearby computer hears Bluetooth advertisements and the browser shows weight and body composition. No Xiaomi account and no vendor cloud in that path.
Attribution (read this)
Credit: Bruno Orlandi, @BrOrlandi. The demo post, the earlier “I want to reverse engineer it” note, and the GitHub repo are his.
What was demonstrated
- Own device. Mi Body Composition Scale 2 (model called out by the author as XMTZC05HM in the repo).
- Agent session. He describes one coding-agent prompt and a few minutes for the first demo.
- What the software does. A local dashboard matches a weigh-in to a profile, stores history, and can optionally ask a model he configures for a weekly summary of his numbers. The README credits body-composition formulas to public projects (openScale and related), not to a private exploit.
- How he described the link. In-thread he said the connection is Bluetooth Low Energy and that the scale’s broadcast is straightforward to receive. The README states the scale does not pair; it advertises while someone is standing on it.
- Official artifacts. Repo BrOrlandi/miscale-dashboard (MIT, Python; 46 stars on 2026-10-13). Promo video: YouTube.
Public metrics (snapshot 2026-10-13)
| Metric | Figure | Note |
|---|---|---|
| Views | ~55.6k | Demo post |
| Likes | 1,263 | Demo post |
| Bookmarks | 494 | Demo post |
| Reposts | 34 | Demo post |
| Source | @BrOrlandi | Demo post |
| Repo | miscale-dashboard | 46★ · MIT · Python |
How this relates to REA
REA is the local MCP/CLI that lets a coding agent inspect software you are allowed to analyze and return Evidence plus unknowns. BrOrlandi’s posts are the same lawful shape — own device, agent, a rebuild you can read. If you want that Evidence loop on a target you own, install REA and keep the human in charge of approval.
Install. Paste into your agent, or run it yourself: npx rea-agents@latest setup. Then restart the client. Walkthrough: /docs/install/ · Codex demo: /docs/install-demo/.
Nearby case. @thinkszyg’s toilet BLE remote is another own-device remote on the same wall.
What to copy (authorized hardware only)
- [ ] The scale is yours, or you have written permission
- [ ] You scope the listener to your unit (the repo’s own docs warn that another scale in range can appear)
- [ ] You do not publish secrets, account tokens, or a neighbor’s measurements
- [ ] If you add REA, you ask for Evidence and unknowns — not a vibe-only story
- [ ] Medical claims stay out; bioimpedance is an estimate
Compliance
Lawful RE only: a scale, watch, or app you own or are authorized to analyze. Do not capture a neighbor’s broadcasts, bypass Xiaomi accounts, or treat this card as a decoder recipe. The author is not affiliated with rea.run.
Takeaways
- Credit @BrOrlandi: one-prompt agent app for his own Xiaomi scale, then an MIT repo.
- Local BLE dashboard for a scale he owns, published as an MIT repo.
- Same lawful shape as other own-device cases; install REA when you want Evidence on targets you may analyze.
Related on rea.run
More on-site cases