Everyone else is cleaning Android TVs. He only does Fire TV: how dumbtv carved out a spot in the "AI cleans your TV" wave

In 2026 October 17, developer Muzzammil Yasin Sabuwala (@DevMuzzammil on X, 52k followers) found the Fire TV at home sluggish: menus were slow, most of the home screen was ads, and only 129 MB of free memory was left. He had a coding agent clean it up over ADB, posted "the great cleanup has begun", and got 455k views. The comments asked for the script, so the next day he created the repo and said "just point your agent at this repo". The post's bookmarks (6,735) outnumber its likes (5,937). 19 days later the repo had 1,956 stars and 213 forks. That number ranks about fifth in this wave of projects. What's worth noting is how it found its position: everyone else cleaned Android TVs, and it only did Fire TV.

What it is

dumbtv is a cleanup kit for Amazon Fire TV (streaming sticks, Cube, Fire TV Edition TVs). It doesn't require root, and every step can be undone. Turn on ADB debugging on the TV, clone the repo on your computer, and tell Claude Code, Codex, or Cursor "clean up my Fire TV, the IP is ...". The agent then follows the AGENTS.md in the repo step by step: it runs a read-only check first and tells you what it will change, asks a few questions (keep Prime Video? keep Alexa? replace the home screen?), and confirms before each step.

Results: about 65 Amazon packages are disabled, including ad placements, ACR (screen recognition of what you're watching), telemetry, and A/B experiments. For the few things Amazon won't let you disable, an open-source DNS filter on the TV blocks their telemetry and automatic updates. The home screen is replaced with one the author wrote, 95 KB, with no network permissions, showing only a clock, your apps, and HDMI inputs. If you don't want to use AI, the same steps are also written as shell scripts.

Why now

He wasn't the spark. In October 16, Ryan Doyle (@doooyle, 6k followers) posted "today I hacked my TV with Claude Code", which got 4.5M views. A speed-comparison video the next day got 3.09M views. On day 17, the cleanyourtv author posted "cleaning my parents' TV", a bigger hit. "Vibe-coding your TV" became a topic overnight.

The dumbtv launch-hype post quoted Doyle's tweet, and the repo went up the next day. On Oct 19, TechCrunch ran "People are using AI to make their smart TVs dumb again," using only two repos as examples: cleanyourtv and dumbtv. It opened with dumbtv's slogan, "Your TV should only do the thing you paid for." The author also gave an interview, saying it was built with DeepSeek V4.1 Flash plus Claude Opus / Sonnet 5.5. The Verge, TechRadar, Tom's Guide, Gizmodo, and XDA followed. The repo's press list shows 150+ articles across 30+ countries.

Stars followed the press coverage:

Where users come from

Mainly three channels, ranked by weight:

The author's own X account. The two posts together got 800k views. The launch post got more bookmarks than likes, so people were saving it to come back and follow the steps, not just scrolling past.

Media. After TechCrunch named it, most follow-up coverage relayed TechCrunch, and the name kept getting carried forward. The author also promoted it: the README header shows logos from TechCrunch, TechRadar and other outlets, and the site has a press contact. I didn't see a Hacker News post.

Short video and tutorials. On TikTok, @nelsonxlee posted a "how to get the TV remote debloat tutorial" video with 99k views, and also wrote a dumbtv tutorial on Substack. Most other big-account videos covered cleanyourtv or just talked about "Claude hacking TVs" in general.

Once stars climbed, what kept people around was the device catalog. Every Fire TV model needs a slightly different package, so the repo asks users to run it and submit their model as a PR. As of 11-06, the repo had roughly 175 issues and PRs, most of them "catalog: [model]" entries, with 30 contributors. The desktop installer had about 7,000 cumulative downloads across 9 releases (GitHub, 11-06). Treat the 7,000 as a floor on installs, since some people built from source.

Why it got a slice

Everyone else in the same wave went after Android TV, while dumbtv was the only one built specifically for Fire OS. Fire TV is the platform with the most ads, and Amazon fights third-party home screens hardest there: newer Fire OS versions grab the HOME button back for Amazon's own launcher, and several of the repo's first-week commits were about working around that. It shipped adaptations for Fire OS 6, 7 and 8, wrote up each package, and added a separate Prime Video toggle (if you watch Prime, its video servers aren't blocked). Anyone searching "Fire TV" finds it, and TechCrunch only listed it separately because it was the Fire TV version.

The second reason was that the "safety" angle was easy to write about. At the time, the press was running stories about AI edits bricking TVs. TechCrunch's closing lines mention someone who bricked an Android TV and someone who got stuck halfway through a Fire TV mod. dumbtv's read-only preflight check, per-step confirmation, list of packages that must not be stopped, and one-click restore gave reporters a "safe version" they could write about.

But don't credit it all to differentiation. WB2024/firestrip, a Fire TV ADB debloat tool doing the same thing, launched five months earlier in May and only reached 80 stars. cleanyourtv, which covers all Android TVs, has 27x as many stars. Focusing on Fire TV gave dumbtv its own niche, but most of the stars came from the topic, the author's following, and press pickup.

Who it's competing with

Since Oct 10, about 40 repos were created with TV or launcher keywords and passed 200 stars. Some people argue you don't need a repo at all: Doyle himself just asked Claude to make the changes, and PCMag and How-To Geek both wrote "you don't need AI" pieces.

Is there room left in this wave?

The peak has passed. dumbtv's weekly star growth dropped from 493 to 204, and single-day gains are now between 20 and 30. Android TV, Fire TV, Samsung, LG and Chinese TV brands all have dedicated versions by now, and major tech outlets in several countries have covered the topic. It is also expanding: the device catalog already includes non-Amazon Android devices like Chromecast with Google TV and Hisense Google TV.

One unverified variable: Amazon's newer Fire TV hardware is starting to move to Vega OS, which isn't Android-based. If the main models migrate, the ADB approach may stop working, and Fire OS-specific tools would lose ground (inference).

If it were you

Bottom line: The TV topic has no room left. What's worth learning is its two moves: first, pick the slice inside a hot trend that the big projects haven't covered; second, turn "AI breaking things" into "safe and reversible." Use both the next time a trend breaks.

• Who it fits: developers with tens of thousands of real followers on X or elsewhere, who own a device the trend hasn't covered and can ship a working repo the same day.
• Who it doesn't fit: people without their own distribution. For the same feature, firestrip got only 80 stars in its first five months. A large share of dumbtv's stars came from the author's 52k-follower account and the TechCrunch mention, and followers don't get those.
• How to do it: When a trend starts, first check which platforms and models the big projects already cover, then pick the leftover slice with the most users and the most frustration (this time, Fire TV). Structure the repo like dumbtv: an AGENTS.md for agents to read, a read-only preflight check, per-step confirmation, a do-not-touch list, and one-click restore, plus a script for people who don't use AI. Then open a device catalog and ask users to submit their model as a PR.
• How to get users: Tease the launch by quoting the viral post, then ship the next day. Write the launch post so it reads as a single line people can follow: "point your agent at this repo." Put a slogan at the top of the README that a reporter can quote directly. When press picks it up, link it in the README, and make yourself available for interviews.
• Timing: within 48 hours of the trend starting. The projects created in the same 48-hour window as dumbtv all caught this wave; repos that came a week later mostly stalled at a few hundred stars.
• Signs you misread it: If nobody in the press or large accounts mentions your slice within 48 hours, or if a major project covers the platform within a week, the slice isn't big enough. Stop putting resources into it and go back to building a tool you can maintain long-term.