Unburden: What If AI Helped You Carry Less?

A developer built Unburden, a browser-based AI task planner that runs the open-weight Xenova/flan-t5-small model locally via Transformers.js, requiring no API key or inference server. The app prompts the model to generate a small first action plus optional next steps, adjusts prompts to the user's selected energy level, and falls back to a clearly labeled rule-based plan if model loading or generation fails. The project is open source on GitHub and built with plain HTML, CSS, and JavaScript.

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady https://dev.to/challenges/hacktoberfest-week3-2026-10-19 . "What if the most helpful productivity tool didn't ask you to do more? What if it helped you carry less?" We've all had days when our to-do list feels less like a plan and more like a browser with 47 tabs open. Reply to that message. Finish that assignment. Buy groceries. Start the project. Remember the other thing you forgot to write down. The problem isn't always a lack of discipline. Sometimes, the hardest part is simply deciding where to begin. That's why I built " Unburden — Do Less, Move Gently https://iqrakhan2909.github.io/Unburden/ ". Unburden is a gentle, AI-powered task planner built around one idea: "reduce the weight of starting, not increase the pressure to finish." Here's how it works: For example, “Finish my DBMS notes, reply to my professor, clean my desk, and start my project report” shouldn't have to become another overwhelming checklist. The goal is to make the first action feel small enough to begin. Unburden is for students, overwhelmed beginners, busy people, and anyone who has ever confused being productive with being constantly busy. I wanted to build something that treats energy as a real constraint, not a character flaw. Try Unburden here: https://iqrakhan2909.github.io/Unburden/ https://iqrakhan2909.github.io/Unburden/ A quick way to explore it: The first model load requires an internet connection. After that, browser caching may help, but fully offline operation is not guaranteed. The complete project is open for inspection on GitHub: " https://github.com/iqrakhan2909/Unburden <https://github.com/iqrakhan2909/Unburden " The app is built with three core files: index.html — the structure and interface. style.css — the responsive visual design. script.js — model loading, inference, task-plan parsing, interaction, and fallback handling. I deliberately kept the implementation small and browser-based. There is no custom backend or hosted AI inference API in the app. I built Unburden using HTML, CSS, JavaScript, Transformers.js, and the open-weight Xenova/flan-t5-small model . The technical decision that matters most is where the inference happens. Instead of sending the user's task dump to a hosted AI chat API, the browser loads the model through Transformers.js and runs inference in the browser. The app doesn't require an AI API key or a custom inference server. The model receives a prompt that asks it to produce five short lines: one very small first action followed by optional next actions. The prompt also changes according to the user's selected energy level. The generated text is parsed into a simple plan, then rendered as an interactive list with checkboxes and a copy option. I also added a fallback path. If model loading or generation fails, Unburden can offer a basic rule-based plan, clearly labelled as a fallback rather than pretending it came from the AI model. That distinction matters: a fallback can keep the interface useful, but it is not a substitute for successful model inference. The design was intentional, too. Soft colours, generous spacing, a little sloth-inspired personality, and language without productivity guilt all support the same goal: make starting feel lighter. The result is a small, self-contained experiment in building AI that respects a person's capacity instead of demanding more of it. For me, open innovation is about more than making source code publicly visible. It's about giving builders and users more control over the technology they depend on. Unburden is a small example of what that control can look like. Using an open-weight model means the model's weights are available to work with under its applicable licence. Developers can inspect the implementation around it, experiment with different compatible models, and adapt the experience without being locked into one hosted AI provider. Unburden runs model inference in the browser. Its task-planning flow doesn't send the brain dump to a hosted AI chat API. That reduces one important source of data exposure and makes the privacy design easier to explain. This isn't a claim that all browser activity is private or that the app is suitable for highly sensitive information. The model and application code still have to be downloaded, and the first model download requires internet access. A beginner can build and publish an AI-powered application without setting up an inference server or managing a paid AI API key. A static website can host the interface while the browser handles inference. That lowers the barrier to experimenting with AI. It also makes the project easier for other developers to study, fork, and improve. Many productivity products are built around doing more, faster. By making the model and surrounding implementation accessible, builders can experiment with different prompts, interaction patterns, and values. In Unburden, that means asking a different question: not “How can AI help someone finish everything?” but “How can AI help someone find one manageable place to begin?” Open innovation doesn't automatically make a model accurate, private, or free of limitations. It gives us more room to inspect, test, and improve the choices we make. And that room to experiment is exactly what made this project possible. The payoff isn't a claim that Unburden saves a guaranteed number of minutes. I haven't conducted a timed user study, and I don't want to invent a productivity statistic just to make the project sound impressive. The intended saving is the effort spent deciding where to begin. Instead of manually sorting a messy brain dump, the user gets a suggested first action and optional next steps. Choosing an energy level helps shape those suggestions, while the fallback means a basic plan may still be available if the model cannot load. The bigger idea is that 'you don't have to turn every unfinished task into an emergency, and you don't have to finish everything to make progress'. Sometimes, the most useful output from AI isn't another enormous plan. It's one small next step — and permission to stop there. I built Unburden for the days when the list is long, the energy is low, and even deciding what to do feels like work. The sloth has a lesson for us: moving slowly isn't the same as standing still. If AI can help us remove a little friction without adding more pressure, that's a kind of productivity worth building for. "Less overwhelm. One next move. Slow is still a speed."