Sloth Mode：少做一點，把事情完成
Sloth Mode - Do less. Get it done.
Sloth Mode 是在用戶電腦本機運行的 AI 助手，可將電郵或訊息草擬回覆、把長文整理成摘要與待辦事項、規劃清單，或為有問題的程式碼提供修正建議。它使用 Ollama 執行開放權重模型 Llama 3.2，毋須註冊或 API 金鑰，資料不會離開用戶的裝置；預設模型為 llama3.2:1b。
What I built
Sloth Mode is a lazy little assistant that does the boring work for you — entirely on your own laptop. Built for Hacktoberfest 2026, Week 3 — Slow and Steady.
Paste anything messy — an email you don't want to reply to, a giant chat thread, a brain-dump to-do list, or buggy code. A local AI model figures out what kind of thing it is, then does the work:
| You paste | You get |
|---|---|
| An email / message | A ready-to-send reply draft |
| A long text / chat thread | A short summary + action items |
| A messy to-do list | A clean, prioritised plan |
| Code with a bug | A suggested fix, explained in one line |
Every finished task tells you roughly how many minutes you saved — and awards Sloth XP (10 XP per minute):
🦥 Baby Sloth · 0 XP
🛋️ Couch Sloth · 200 XP
🧘 Zen Sloth · 600 XP
👑 Legendary Sloth · 1,500 XP
No sign-up, no API keys, no data leaving your machine. The AI is Ollama running the open-weight Llama 3.2 model, and your history lives in the Lazy Vault — a drawer in your browser's localStorage. One button wipes it all.
Demo
🎥 Demo video: [attach the 50-second demo video here]
🔗 Live showcase: https://mohidshaikh.tech
💻 Code: https://github.com/captaincpr9-bot/Sloth-Ai
The showcase page is just that — a showcase. The app itself runs on your computer, not on a server, so there is no hosted demo to try here. The sloth would never make you upload your emails to the cloud.
How it works
- You paste — anything messy, no task type to pick, no prompt to write.
- The sloth classifies — the local model works out what kind of thing it is: reply, summary, plan, or fix.
- The work is done — it runs the matching task, shows your time saved and XP, and files the result in the Vault.
The minutes-saved number is an honest rough estimate, not a measurement.
Built with Python + Flask, an embedded web frontend, and Ollama + Llama 3.2 (the default model is llama3.2:1b, so it even runs on a low-RAM machine). During development I also tested the full 3B model through a Colab GPU bridge — but the app itself ships 100% local.
Run it yourself
You need Python 3.10+ and Ollama. Then:
來源：DEV Community · dev.to