Example note06:41
The B-side elevator is stuck; the super looks before 10.
Potluck moved to Sunday at 4, and hot water is off Thursday morning.
Slow and steady · local Gemma
Leave the chaton the nightstand.
Nightbrief reads the overnight pile — a group chat, a mailbox export, a text file — and leaves two lines for the morning. Gemma runs in this browser, on your GPU. The messages do not go anywhere.
Paste it before bed
Drop a .txt or .md export, or paste the thread. Slack and Gmail stay where they are — this version wants the text you already have.
Let Gemma read
WebLLM loads an open Gemma build with WebGPU. The first run downloads weights into this browser. Later mornings use the cache.
Two lines, then coffee
Decisions and time changes come first. Anything that can wait folds into a short list you can ignore.
Nothing phones home with your messages.
- Open-weight Gemma. The studio defaults to Gemma 2 2B, with Gemma 3 1B when you want a lighter load.
- Local WebGPU. Inference stays in the page. There is no Nightbrief server and no API key.
- Your text stays put. The paste is not uploaded. The only download is the model, from Hugging Face, until the browser cache is warm.
- A week for slow tools. Built during Hacktoberfest Week 3, October 19–25, 2026. Read the challenge.