VictorTran1023/android-tv-toolkit

This is not a polished app but a documented operating procedure plus helper scripts for people who want to reclaim an Android TV from vendor preloads without root or a firmware flash. If you already SSH into your set-top hardware and enjoy reading logs, it can save you hours of trial and error. If you just want a faster TV, the multi-step gate-and-regression discipline is more ceremony than you likely want.

The problem

You want to stop the home screen from showing ads, stop a voice assistant from holding a wake lock, install a YouTube client that works without Google services, and know whether you made things better or broke the remote. On stock hardware the friction is spread across four points: adb unauthorized with no on-screen prompt, package verification blocking sideloads, missing GMS, and vendor daemons that reappear after a reboot. The project turns that into a repeatable sequence: identify the device, authorize, take a snapshot, disable packages one at a time, reboot, then run a scripted regression check. The result is a changed set with a recorded baseline you can return to.

What's different

Most of the value is in the discipline rather than the scripts. Before changing anything the toolkit saves package lists, disabled packages, three settings tables, system properties, and the current launcher, screensaver and animation values. Every optimization is then performed one step at a time, with a reboot and a regress.ps1 pass that probes the live launcher and settings packages instead of hardcoding a vendor name. That gate-then-check loop is the mechanism that separates this from a one-line debloat script you found in a forum thread.

The second distinctive piece is the HUD, a roughly 41 KB native Java app with no third-party dependencies that shows CPU, memory, VPN state, a temperature reading pulled via dumpsys thermalservice, and local weather as a screensaver. It does not take remote focus and fades after ten seconds. It is also honest about a gap: GPU usage is listed as unavailable because the vendor node refuses access. The tvOS-style launcher is explicitly a design study and interaction prototype, not installed and not taking over HOME, so do not adopt this expecting a new home screen.

Who should try it

Good fit: a TCL/Thunderbird owner who has seen the T7K case and wants the same method with commands, evidence and rollback notes. Also a reasonable fit for someone with a Xiaomi, Sony or Hisense box who is willing to treat the brand table as a starting hypothesis rather than a guarantee. Poor fit: a user who wants a click-to-clean tool, or who owns a HarmonyOS, LG webOS or Samsung Tizen device, which the README says are not Android TV and are unsupported.

Requirements and limitations

You need Android TV 9 or newer for the scripts, and Android 10 or newer if you want the HUD. The computer side expects adb and PowerShell for the preflight/regression scripts, with a separate Python 3.9+ helper for the legacy public-key authorization case. Both are documented in the usage guide. The setup expects the TV and computer on the same subnet, and the first connection needs a debug-authorization prompt on the TV.

The larger limitation is coverage. The README states that TCL is the only fully tested brand, while other brands are marked verified, reference or unverified in the brand table. The documented CPU and temperature improvements (48% to 9% CPU, roughly 10 °C cooler) are a single TCL T7K sample, not a promise for other models. Sideload paths differ by brand, some sets need a USB stick or a vendor app store, and the author notes a specific trap: after pm disable-user you must wait at least 60 seconds before rebooting or PackageManager may restore the packages. Reversibility is real, but restoring factory settings to recover a lost feature costs your data.

HotRepo's take

HotRepo first recorded this repository on 2026-10-24, and on 2026-10-25 it appeared at #18 on the new-repository board for Java with 81 total stars. That is a two-day sample and not a daily growth series, so persistence cannot yet be judged. Before relying on it beyond TCL, check whether your exact model is marked verified or only reference in the brand documentation.