Borrowed Recipe
A pixel hero with a new water nag every break
Break reminders you won’t learn to tune out, with a new nag every time, written on your own computer.
Install Python and Ollama, copy the Break Buddy app, pull one small Gemma model, and a pixel hero pops up to tell you to drink water, rest your eyes or stretch, in the tone you pick.
A big thank-you to Kumuda Sri P for building it, sharing the code and writing up what broke along the way; you only copy commands and type one sentence.
🧾 What’s cooking
- Why: by their account, reminder apps repeat the same sentence until your brain learns to skip it. Their fix is a reminder that’s worded differently every time.
- The dish: Break Buddy, a Windows app that runs in the background. When a break is due, a pixel-art hero points at an RPG-style message box and types out a fresh message (Break Buddy README).
- What it does, per the README:
- you set it up by typing one sentence that lists each reminder, how often it comes, and the tone you want
- a tone of your choice (sarcastic friend, calm coach, pirate and so on)
- it holds reminders during Zoom, Google Meet, Teams and Webex windows, then shows one combined popup after the call
- a nudge to take a proper break after 2 hours of continuous computer time
- Snooze (10 minutes) and Done on every popup, plus a tray icon to pause, resume or quit
- How setup reads your sentence: by their account, Pydantic and Ollama’s structured output force setup to return valid JSON, which is checked and retried up to 3 times. The README adds that it runs at temperature 0 with thinking off and shows you the result before saving.
- Secret ingredient: the model runs on your machine through Ollama, so window titles never leave it, per the README. The message writer only sees a label like “coding” and a number of minutes, never the title itself.
- It’s their entry in DEV’s Hacktoberfest Open-Source AI Challenge, Week 3, whose “Slow and Steady” theme asks for open-weight builds that reward laziness (DEV: Hacktoberfest Open-Source AI Challenge, Week 3).
🍳 Difficulty · ⏱️ Prep time
| No coding | |
|---|---|
| Difficulty | 🟡 Comfortable copying commands into a terminal |
| Prep | About 30 minutes, mostly downloads; the author puts clone to first popup at about five minutes |
🛒 Ingredients
- A Windows PC. The README says Windows only, Windows 10 or 11. Ollama itself needs Windows 10 22H2 or newer, Home or Pro (Ollama docs: Windows).
- Python 3.10 or newer, as the README asks. Python’s own guide points Windows users to the Python install manager, from the python.org downloads page or the Microsoft Store (Python docs: using Python on Windows).
- Ollama for Windows (Ollama: download for Windows). It installs in your account without administrator rights (Ollama docs: Windows). Ollama added Gemma 4 in v0.20.0 (Ollama release notes: v0.20.0), so update an older install.
- Git for Windows, for the README’s
git clonecommand (Git: install for Windows). - Disk space: at least 4GB for Ollama itself, plus room for models (Ollama docs: Windows). The
gemma4:e4bmodel is listed at 6.6GB to 9.5GB (Ollama: gemma4 model page).
👩🍳 Method: no coding, just commands
- Install Python, Git and Ollama from the Ingredients links. Once installed, Ollama runs in the background and the
ollamacommand works in Command Prompt or PowerShell (Ollama docs: Windows). - Copy the app with the README’s first two commands below:
git clonedownloads the Break Buddy repo, andcdsteps into its folder. - The next two make a private Python space for the app and switch it on, then
pip installadds the app’s parts. Python recommends a separate space like this for each project (Python docs: using Python on Windows). In PowerShell the switch-on script isActivate.ps1, and you may need to allow scripts first (Python docs: venv). ollama pull gemma4:e4bdownloads the model, the app’s default (README).python setup.asks what reminders you want. Type one sentence; it prints what it understood, and you answerpy yto save it (README). The first run is slow while the model loads.python breakbuddy.starts the app. A blue icon appears near the clock; check the ^ arrow if you don’t see it (README). The commands are below, in the README’s order.py
git clone https://github.com/kumudasrip/Break-Buddy.git
cd Break-Buddy
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
ollama pull gemma4:e4b
python setup.py
python breakbuddy.py
Two of the README’s example sentences, to paste at the setup prompt:
remind me to stand up every 30 min, pirate voice
drink water every hour and a half, angry gym bro
Want it to start with Windows? The README’s own last step: press Win+R, type shell:startup, add a shortcut to the app’s pythonw.exe and breakbuddy.py, then set Start in to the app folder. The README says the app can’t find your settings without that last part.
😋 Taste test
- Quick try: set
TEST_near the top ofMODE = True breakbuddy.py, and minutes become seconds. Set it back toFalsefor real use (README). - By their account, the small model was enough: it matched or beat the bigger ones on window labels and message quality, and caught all 7 test meetings with no false alarms.
- Generic messages like “Time to stretch!” mean Ollama isn’t running or the model isn’t pulled (README troubleshooting).
- Misread sentence? Answer
nand rephrase (README).
🔥 Watch the stove
- Windows only (README).
- Meeting detection reads window titles. The README says Zoom’s home screen also holds reminders, and a call’s short “Connecting…” screen can let one slip through.
- Idle time counts as a break, so a long video without touching the keyboard resets the screen-time counter. Reminders still pop up while you’re away and close after 25 seconds (README).
- A solid magenta popup background? Set
USE_inHERO = False breakbuddy.pyfor a plain popup (README). - Their own caveat: their model test was small (10 sentences, 36 window titles), and they wrote the correct answers themselves.
📎 Where this came from
- Original recipe: Kumuda Sri P on DEV: "I Built a Pixel Hero That Nags Me to Drink Water" (Oct 23, 2026)
- Break Buddy on GitHub
- Ollama docs: Windows
- Ollama: gemma4 model page
- Ollama release notes: v0.20.0
- Python docs: using Python on Windows
- Python docs: venv
- Git: install for Windows
- DEV: Hacktoberfest Open-Source AI Challenge, Week 3