This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady I built the Cat Break Agent — a tiny local AI that watches you rot in your chair, gently roasts you, and then force-feeds you a fresh cat video like a digital mom who also has main-character energy. Every hou
r (or every 30 seconds when I’m testing and feel chaotic), Gemma (running locally through Ollama) looks at the clock and how long you’ve been glued to the screen. It picks a vibe — calm, chaotic, or cute — writes a short, slightly unhinged nudge to stretch and drink water, then yeets a matching recent YouTube cat video into your browser while pinging a desktop notification. Who gets to be lazier? Anyone who has ever opened “just one more tab” and emerged three hours later questioning their life choices.
Now the AI does the decision fatigue for you. You’re welcome. Run it locally (Windows + Ollama + a free YouTube Data API key). Repo: https://github.com/Syed-rafath/breakitty (Insert your screen recording / GIF of the notification popping + browser auto-opening a chaotic kitten here.
The 30-second test mode is comedy gold.) The whole thing is one Python file that feels like it was written by someone who has both ADHD and a soft spot for orange cats. Key bit — the local model call: prompt = f"""It is {clock}. The user has been at their computer for {hours_worked:.1f} hours. Choose the best cat video vibe from: calm, chaotic, cute.
Write a short, funny, gentle nudge to stretch or drink water. Reply only as JSON: {{"vibe": "...", "message": "..."}}""" It has caching, rate-limit protection (so you don’t get rate-limited by Google while trying to be healthy), and a fallback message for when the 1B model decides to freestyle instead of outputting valid JSON. Because even open-weight models have their chaotic days. Open-weight model: gemma3:1b (and later compared with 4b) via Ollama.
Fully local. No API bills. No “your prompt has been sent to the mothership.” Glue: Plain Python, plyer for notifications, a little YouTube Data API v3 for fresh videos (last 30 days only), and a local JSON cache so it doesn’t spam Google every break. The model only decides the vibe and writes the nudge.
Everything else is deterministic code. Classic “AI does the fun judgment call, Python does the boring reliable stuff” architecture. Built on a laptop with 8 GB RAM and the attention span of a cat watching a laser pointer. Perfect.
Because if this ran on a closed API, every time I told the model “I’ve been sitting for 4.7 hours,” that sad fact would leave my machine. With open weights + Ollama, the only thing that leaves is a generic search query like “funny playful cat video.” My shame stays local. The cats stay public. Open weights made it possible to ship something that feels personal and private on potato hardware, instead of another “just sign up for our SaaS” productivity guilt machine.
Also, Best Use of Gemma category? I’m looking at you. I actually used it this week while writing this post. Result: three stretches, two glasses of water, and one video of a cat falling off a couch that made me laugh so hard I almost spilled the water.
Net productivity gain: debatable. Net serotonin gain: undeniable. This is the spiritual opposite of those “touch grass” apps. This is “touch cat video, then maybe stretch.” Peak 2026.
Best Use of Gemma (obviously) Overall Slow and Steady excellence (a cat can dream) Built with open weights, local inference, and the eternal knowledge that the internet was made for cats. Now if you’ll excuse me, my agent just told me it’s time for a chaotic break. devchallenge #hf26challenge #hacktoberfest #productivity #opensourceai #gemma
Key Takeaways #
- •It picks a vibe — calm, chaotic, or cute — writes a short, slightly unhinged nudge to stretch and drink water, then yeets a matching recent YouTube cat video into your browser while pinging a desktop notification.
- •This story was reported by Dev.to , covering developments in thedev space.
- •AI advancements continue to reshape industries — read the full article on Dev.to for complete coverage.
📖 Continue reading the full article: