Overnight Brain: A Lazy AI Assistant That Works the Night Shift

Ciara Reed — Sat, 24 Oct 2026 22:09:27 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

What does it do?

Overnight Brain is a lazy person's AI assistant. During the day, I drop things into a queue: questions I've been meaning to look into, long articles to summarize, big goals to break into steps, and messages I keep putting off. At night, Google's open-weight Gemma model runs right on my laptop and works through the list one task at a time while I sleep. In the morning, I wake up to a digest with everything done.

What effort does it save?

It saves me from sitting in front of a chatbot waiting for answers, and from the mental load of carrying around a to-do list of "I'll look that up later." I hand it off once and stop thinking about it.

Who gets to be a little lazier?

Anyone with a full day and a tired brain by evening. I start work at 6 a.m., so my evenings are filled with exercising, dinner, and other tasks. Now I can queue things up before bed and read the answers with breakfast.

Demo

I queued three tasks at 10 p.m., went to bed, and at 1 a.m. my laptop started working through them on its own.

There’s no deployed link because Overnight Brain is designed to run privately on your own computer. Here’s what a real overnight run looked like on my MacBook.

The worker rested until 1 a.m., finished all three tasks, and saved the digest at 1:04 a.m.

What I woke up to: a plan for this post, a way to explain open-weight models, and a LinkedIn draft. (Click to enlarge.)

Code

creed725 / overnight-brain

A lazy AI assistant that runs Gemma locally overnight and leaves you a morning digest.

Overnight Brain 🌙

Hand it off tonight. Read it with your coffee.

Overnight Brain is a lazy person's AI assistant. During the day you drop questions, long articles, big goals, and messages you keep putting off into a queue. At night, a small open-weight model running on your own laptop works through them one at a time, slowly, while you sleep. In the morning you get a digest.

Built for the Hacktoberfest 2026 Open-Source AI Challenge, Week 3: Slow and Steady.

Why slow on purpose

Most AI tools race to answer in seconds, which means a giant model in someone else's data center. Overnight Brain flips that. If the answer can wait until morning, a small open-weight model like Gemma on an ordinary laptop is plenty. It can take its time, run with fewer threads, rest between tasks, and nobody has to watch a spinner.

Why open matters here

…

View on GitHub

How I Built It

I built Overnight Brain with help from an AI assistant. I brought the idea, and it wrote most of the code. My job was setting it up on my own Mac, testing it, troubleshooting, and publishing it.

Open-source AI used: Gemma 4 E2B, an open-weight model from Google, running locally through Ollama, an open-source tool for local inference. No cloud AI services are involved.

The pieces:

Gemma 4 (E2B): Google's open-weight model, the "brain" that writes the answers
Ollama: A free app that runs Gemma on my laptop
Python with FastAPI: The small web app where I add tasks and read the digest
SQLite: A tiny database that holds the queue overnight
A worker script: Waits until 1 a.m., then works through the tasks one at a time

At 1 a.m., the worker sends each task to Gemma through Ollama on my own laptop, saves the answer, and then has Gemma write the short good-morning note at the top of the digest.

The 8 GB plot twist: My first model download was 9.5 GB. Then I checked my MacBook and found it only has 8 GB of memory, so the model would never fit. I switched to a smaller, compressed version of Gemma that's built for lighter machines, and it ran fine at about a minute per task.

Gentle mode: Since nobody's waiting on the answers, the worker uses fewer CPU threads and rests between tasks, so my laptop stays cool and quiet overnight.

Why Does Open Innovation Matter?

I work in federal payroll, where protecting sensitive information is part of my everyday job. That shapes how I think about AI. Most AI tools send whatever you type to someone else's servers. With an open-weight model like Gemma running on my own laptop, nothing leaves my computer. I can hand off personal notes, half-formed ideas, and drafts without wondering where they end up.

Open models also make this kind of project possible for everyday people:

It's free. No subscription, no API key, no per-question bill. My laptop was going to sit on my desk overnight anyway.
It runs on what I already have. A MacBook with 8 GB of memory was enough, because an open model came in a smaller size built for machines like mine.
It's mine to change. Anyone can swap in a different model, rewrite the prompts, or add new kinds of tasks.

Slow and steady only works because the model is mine. A closed AI service is built to answer fast. An open model on my own laptop can take its time while I sleep.

The Lazy Payoff

I used it this week. On Wednesday night, I queued three tasks around 10 p.m.: a plan for writing this post, a question about explaining open-weight models, and a LinkedIn announcement. Then I went to bed. At 1 a.m., my laptop started on its own and finished all three in about four minutes. When I woke up at 6, the digest was waiting.

The effort it saves isn't really the minutes. It's that I didn't sit in front of a screen waiting, and I didn't carry those to-dos around in my head all evening.

Honest lessons from a small model:

It can overstate. My good-morning note said my DEV post was "finalized" and "ready to go." It had only written a plan.
Check its facts. It said open-weight models let you "see exactly how the model makes decisions." Not quite. Open weights let you download, run, and customize a model, but they don't explain why it gives a particular answer.
Expect templates, not finished work. The LinkedIn draft came with blanks like "[Link to project/GitHub]" for me to fill in.

Overnight Brain is a great head start, not a replacement for my own judgment.

Try It Yourself

I'd love for you to try Overnight Brain on your own laptop. Setup takes about 15 minutes, plus download time. Just follow the README in the GitHub repo linked in the Code section above.

You'll need Python, the free Ollama app, and about 5 GB of free space for the Gemma model. I built and tested it on a Mac, so if you try it on Windows or Linux, let me know how it goes.

If you do try it, tell me in the comments:

What did you hand off before bed?
How long did your laptop take to finish?
What kind of task should it learn next?

Feedback, ideas, and pull requests are all welcome!

Prize Categories

Gemma

Greetings and happy hacking! 👩🏽‍💻

Ciara Reed — Sat, 24 Oct 2026 19:04:01 +0000

DEV Community: Ciara Reed