Skip to main content
Back to trending

dsh-see-now

Lets AI see your screen in real time: filters frames by amount of change and stitches multiple frames into one image, so static screens cost almost no tokens.

Translated from Chinese · show original

让 AI 实时看见你的屏幕:按变化量筛选帧、多帧拼进一张图,静止画面几乎不花 token

  • ai-agent
  • dsh
  • dsh-plugin
  • ffmpeg
  • llm
  • realtime
  • screen-capture
  • token-efficient
  • vision
View on GitHub
Stars
164
Forks
8
+ today
—
Created
2d

Ranking data as of November 3, 2026 (UTC).

Overview

Lets AI see your screen in real time: filters frames by amount of change and stitches multiple frames into one image, so static screens cost almost no tokens.

The project is written in JavaScript and has 8 forks. It was created 2d ago.

AI-generated summary of the repository README — may contain errors.

What it is

`SAOS-3624/dsh-see-now` ("See Now") is a JavaScript plugin for DSH, licensed under Apache-2.0. Its description (in Chinese) says it lets an AI watch your screen in real time while spending almost no tokens on a static picture.

Capabilities described

- Each captured frame gets a change score from 0 to 255, computed from weighted HSV differences on a 64×36 grid. The README summary gives roughly 0.2 ms per frame. - A frame is kept only when the screen changes abruptly or keeps moving until enough change accumulates. - Kept frames are combined into one image, spreading its fixed cost across 4 to 8 frames. - Topics include ffmpeg, screen-capture, vision, llm and token-efficient.

Getting started

The README summary says to restart DSH after installing, since the plugin tree is composed at process start. It also mentions a web profile for browsers and a `--profile desktop` option, which needs the desktop app to have been opened once. Exact install commands were not supplied, so follow the README.

Installation
git clone https://github.com/SAOS-3624/dsh-see-now.git
cd dsh-see-now
# see README for setup

Frequently Asked Questions

What problem does See Now solve?

Per its description, it lets an AI see your screen in real time while keeping token use low, because it sends only frames where the picture changed.

How does it decide which frames to keep?

Each frame gets a 0-255 change score from HSV differences on a 64×36 grid. Frames are kept on an abrupt scene change or after enough accumulated motion.

Do I need to restart anything after installing?

The README summary says to restart DSH so the plugin takes effect. Install commands were not supplied, so check the README.

What license and language does it use?

JavaScript, under the Apache-2.0 license.