LeaveNote

Stop coding when you want. Paste your changes, and a local AI writes the handover note that helps tomorrow-you start where you left off.

The AI runs on your device using Qwen2.5 (open-weight) and WebLLM. First load downloads model files (hundreds of MB). Requires WebGPU; devices without it get an honest message, not a fake demo. No API key, no account, no server inference.

Check device support by loading the model.

Project name (optional) What were you working on? (optional, one line) Your code changes (paste git diff output)

0 / 1600 characters read - longer diffs are trimmed, paste the important part

No note yet. Your handover appears here, fully written on your device.

Diagnostics (stays in this browser)

Last saved note

No AI on this device? Use the manual template

Same four questions, answered by you instead of the model. Copy it and fill it in anywhere.

LEAVENOTE - [project] - [date]
Working on: [one line]

WHAT CHANGED:
- [what you changed, naming the file or function]
UNFINISHED:
- [what is half-done]
START TOMORROW: [the single first thing to do when you sit back down]
WATCH OUT:
- [leftover TODOs, debug lines, anything easy to forget]

Privacy & limits

Your diff and notes are processed by a model running in this browser, not an AI service. The app has no server, no analytics and no uploads. Hosting/CDN/model providers receive ordinary file requests and your IP address. Model files may stay in the browser cache; saved notes live in this browser's local storage until you clear them.

A small model can misread a diff or invent a next step. The note is a memory aid, not ground truth - review it, and keep your real source of truth in git. Long diffs are trimmed to the first 1600 characters, so paste the part tomorrow-you most needs.