You record, you put in the effort, and the video goes nowhere. Meanwhile, a small channel on the same topic gets hundreds of times its usual views. The trending radar shows you, before you record, what is growing and whether it grew because of the topic (you can use the idea) or just because the channel is big (you can't).

youtube-bombando is a small system that runs by itself on your computer every morning. It searches YouTube for the week's videos on the topics you choose, records each one's views for 7 days and builds a panel showing what is growing fastest. For each video, it compares the views with the channel's own usual numbers and tells you whether the signal comes from the content, from who posts (a big channel) or from who speaks (a guest). It's for people who make videos or want to start a channel. All it needs is Python and a free tool that reads YouTube; no key, no paid API.
Watch the project explained (in Portuguese). Prefer vertical? Watch the Short
At 7:15 AM the collection finishes and the panel updates by itself. You pick the topic at the top and the order of the videos; each sort option is explained below the selector.




| Sort by | What it means |
|---|---|
| Score | The hottest right now: how much faster the video grows than the others on the same topic, plus how far it went beyond the size of the channel. |
| Speed | Views per hour since publication. Favors new videos. |
| 24h gain | Views gained since the previous day's measurement (shows up from the 2nd day of collection). |
| Multiplier | Views ÷ subscribers. Above 1× the video reached more people than the channel has subscribers. |
| × channel | Views ÷ the usual numbers of the channel's own latest videos. 3× or more = burst the channel's bubble. |
| Views | Total views, regardless of age or channel size. |
| Newest | The most recently published first. |
A video with 1 million views on a channel with 5 million subscribers teaches you nothing: the audience was already there. The same million on a channel with 7 thousand subscribers teaches a lot. That's why the panel compares each video with the channel's own usual numbers (the median of its latest videos).
Got 3 times or more the channel's usual. The topic pulled it. This is the idea worth studying and doing your own way.
A big channel (over 500 thousand subscribers) at its usual pace. The audience already existed; copying the topic won't bring the same result.
There's a guest (interview, podcast, talk or a well-known name) and it beat 1.5× the usual. The person pulled it.
It's a signal, not proof of cause: simple math (views ÷ the channel's usual), no AI. When the channel has no baseline yet, the video goes under “No data”, never a guess.
320 videos across 7 topics (artificial intelligence, n8n, claude code, AI agents, vibe coding, making money with AI, chatgpt).
of the views came from videos that got 3× or more their channel's usual.
of the views came from big channels at their normal pace (who posts). On the “artificial intelligence” topic, 31%.
the biggest jump of the day: a small channel got 810 times its usual. A 5-hour course on a channel with ~7 thousand subscribers got 513×.
These numbers change every day; the panel shows today's.
From here down is the technical part: install it, choose the topics and leave it running every day. You don't need this part to understand everything above.
Uses yt-dlp (a free, open tool) to find the week's videos for each topic and read the views, just like a browser would. Pauses between calls and a cap per run.
One snapshot per video per day in a local SQLite database: speed, 24h gain, multiplier, × channel and the score. All explained in the README.
A single index.html that opens in any browser, served on port 8795. It can also show up inside another panel (INEMA shows it in a “Network trends” card).
Each video is tracked for 7 days from when it was found. Each channel's baseline (latest 20 videos) is kept for 7 days. The first full collection took about 20 minutes.
No key, no extra Python package. The automatic schedule uses Linux systemd; on Mac and Windows you run the collection by hand (or with the system scheduler).
python3 --versionInstall it and keep it up to date (YouTube changes and yt-dlp keeps up).
# recent Ubuntu/Debian sudo apt install pipx && pipx install yt-dlp pipx ensurepath # then close and reopen the terminal # other systems python3 -m pip install -U yt-dlp yt-dlp --version
No git: download the ZIP.
# no git? sudo apt install git git clone https://github.com/inematds/youtube-bombando cd youtube-bombando # downloaded the ZIP? the folder is youtube-bombando-main
The tests and a run with sample data, without touching YouTube. The test panel goes to dados/teste/painel/. The expected ending is rodada ok: 7 nichos, 3 novos, 2 fotos, 1 erros: that 1 error is on purpose (a sample private video).
python3 -m unittest discover -s tests python3 bombando/rodada.py --fixtures tests/fixtures # open dados/teste/painel/index.html in your browser
Edit config.json. Each topic has a name, the search term and whether it's active. The "nichos" list comes with 7 AI topics: replace them with yours or set "ativo": false on the ones you don't want. Comma between items, none after the last one; leave the rest of the file as it is.
{
"nichos": [
{ "nome": "easy recipes", "busca": "easy recipe", "ativo": true },
{ "nome": "desserts", "busca": "easy dessert", "ativo": true }
],
…
}
It can take from a few minutes to half an hour, depending on the number of topics. For a single topic, use --nicho.
python3 bombando/rodada.py python3 bombando/rodada.py --nicho 'easy recipes'
The panel is a single file. Open it directly or serve it on port 8795. No server: double-click painel/index.html. It stays empty until the first collection finishes.
python3 -m http.server 8795 --directory painel # http://localhost:8795
Installs the 7:15 AM schedule and the panel server as services for your user. Safe to run again. The computer must be on and you logged in; if it was off at 7:15, the collection runs as soon as you turn it on and log in. The panel is visible to other devices on your home network (port 8795).
bash deploy/instalar.sh systemctl --user list-timers youtube-bombando.timer
To check that the automatic collection ran: journalctl --user -u youtube-bombando. To turn everything off: systemctl --user disable --now youtube-bombando.timer youtube-bombando-painel.service.
Without collecting anything, for example after changing something in the look.
python3 bombando/rodada.py --so-painel