MicroSloth：離線 AI 智能體趁食物加熱清空「Micro-Dread Queue」

What I Built

We all know the awkward 60-to-90-second microwave limbo: waiting for leftovers to heat up or the kettle to boil is too short to sit down and start deep work, so we pull out our phones, start doomscrolling, and lose 20 minutes while our food gets cold.

For International Sloth Day, I built MicroSloth 🍿🦥—a 100% offline "microwave waiting-time" agent that clears out my digital micro-chores while my food spins.

Throughout the day, whenever I encounter a tiny, annoying task I dread doing (drafting a polite text to my landlord about a leaky sink, turning messy brain-dump notes into clean bullets, or looking up a specific git or bash command), I toss a one-liner into MicroSloth's Micro-Dread Queue. When I walk over to heat up food or coffee:

1. I set the Microwave Timer on MicroSloth and click start.
2. The UI switches into Sloth Zen Mode, prompting me to step back from the screen, unclench my jaw, and stretch.
3. Simultaneously, a local open-weight LLM running via Ollama loops through my Dread Queue in the background and finishes every single micro-task before the timer hits zero.
4. At 0s, a pure-Python synthesized microwave DING! 🔔 plays, my food is hot, and all my annoying replies and notes are ready to copy-paste (or download as a single .md report).

Anyone who procrastinates on 2-minute digital errands gets to be guilt-free lazy—turning dead kitchen waiting time into completed chores without lifting a finger.

Demo

Pictured above: MicroSloth auto-detecting local Ollama models, running through the Dread Queue during a microwave cycle, sounding the synthesized bell chime at 0s, and exporting the finished tasks to Markdown.

Code

micro-sloth

GitHub Repository: https://github.com/afreensumai64/micro-sloth

How I Built It

MicroSloth is built entirely around local inference and small, fast open-weight models so that tasks finish faster than a standard 60-second microwave cycle:

• Local Inference Engine: Ollama (ollama-0.6.3 Python SDK) serving models locally over localhost:11434[cite: 1]. The app automatically queries ollama.list() on startup to detect whichever open-weight models are already downloaded on the machine.
• Open-Weight Models: Built for lightweight 3B–7B parameter models like llama3.2:3b, qwen2.5:3b, or mistral. Because 3B open-weight models have near-instant time-to-first-token on consumer laptop hardware, MicroSloth can knock out 3 to 4 distinct tasks in under 25 seconds.
• UI & Timer Harness: Built with Python and Streamlit[cite: 1, 2], pairing a real-time countdown metric with sequential prompt execution and expandable Markdown code blocks for one-click copying.
• Zero-Dependency Audio Synthesizer & Local Persistence: Using Python's built-in wave, math, and struct modules, MicroSloth mathematically synthesizes a dual-tone C7/C8 metallic microwave bell chime (44.1kHz WAV) in memory when the timer hits 0s and auto-saves every batch of finished chores to completed_dread_queue.md[cite: 2].

Why Does Open Innovation Matter?

A "microwave-gap" agent for tiny daily chores only makes sense because of open-weight models and local inference:

1. Zero-Friction, $0.00 Micro-Runs: Nobody wants to wire up credit cards, monitor token billing, or pay per-call API fees just to format three grocery items and draft a two-sentence text message while heating up lunch. Running open weights locally via Ollama means I set it up once and run it 10 times a day for free forever.
2. Complete Brain-Dump Privacy: My "Micro-Dread Queue" holds unfiltered personal thoughts, landlord messages, messy meeting notes, and internal project snippets. With local inference, my raw brain dumps never leave my laptop or pass through third-party cloud servers.
3. Offline & Instant: Whether my Wi-Fi drops or I'm working from a spot with spotty internet, MicroSloth works in full Airplane Mode with zero network latency.

The Lazy Payoff

The biggest cost of tiny tasks isn't the 2 minutes it takes to do them—it's the mental friction of starting and the 20 minutes lost to phone doomscrolling when you procrastinate on them during breaks.

Using MicroSloth this week during coffee and snack breaks cleared out nagging replies and terminal lookups that usually sit on my sticky notes for days. Better yet, because the screen tells me to step back and stretch until the microwave bell rings, I actually get a real 60-second physical break and come back to a finished checklist.