This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
Planio – Goal to Plan maker: type out everything on your plate and get tiny, scheduled steps on a sticky note that shows only today. Runs an open-weight model (Qwen3 8B) on your own computer via Ollama. It helps to plan out bigger things into smaller tasks so that one does not get overwhelmed and get to work fast.
Demo
Check the app here - https://meetpatel2209.github.io/HacktoberFest2026/DevChallenge1/
Uses the local running model - qwen3:8b.
Code
Check the DevChallenge1 here :
HacktoberFest 2026
Projects for the DEV Hacktoberfest 2026 challenges. Each challenge lives in its own folder, with its own README and its own page on GitHub Pages.
| Challenge | What it is | Live | Details |
|---|---|---|---|
| DevChallenge1: Weekend Challenge, Build for a Friend | Planio – Goal to Plan maker: type out everything on your plate and get tiny, scheduled steps on a sticky note that shows only today. Runs an open-weight model (Qwen3 8B) on your own computer via Ollama. | Open the app | README |
Code is MIT licensed.
.
How I Built It
Under the hood, an open-weight model (Qwen3 8B) runs on your own GPU through Ollama. It reads your free time and breaks each goal into concrete steps. Plain, tested JavaScript then checks the answer and schedules the steps into your free evenings. There are no accounts and no backend, and the plan is saved in your browser.
Why Does Open Innovation Matter?
A brain-dump is about as personal as text gets: deadlines you're behind on, things you're avoiding, errands for family. I didn't want any of it sent to an AI company, and with a local open model it isn't.
• Privacy. The model runs on the friend's own GPU. The web page talks only to localhost, and the plan is stored only in their browser.
• Free, forever. No API key, no bill, no account. The friend can use it as much as they like.
• Mine to change. The app talks to the model through one small function. Swapping Qwen3 for another open model was one word on the command line, and an in-browser runtime like WebLLM would slot into the same place.
To be honest about what does touch the network: installing Ollama and downloading the model (once), and loading the page from GitHub Pages. I bundled the handwriting font instead of loading it from Google, so once the page has loaded, planning itself needs no internet.
My Agent Session
Built over one weekend with Claude Code because just wanted to tinker with open weight models and run them locally.
The handoff
My friend said that it created tasks but sometimes it hallucinated and created detailed tasks that were maybe not required.
I accept this as a drawback given the model constraints. It could be further improved by giving the full calendar context maybe, will check!
Note - got to learn about the open weight models, running them locally and using them in an application. It was a fun, small experiment that I completed over weekend. Not the best given the hardware constraints but good for the first cut. Open to any suggestions and changes.
Top comments (0)