See Now is a community plugin for DeepSeek Harness (DSH) that transforms screen, camera, or video file inputs into frame sequences filtered by change magnitude. Each frame receives a change score; only frames with sudden scene switches or accumulated continuous motion are retained, and these selected frames are combined into a single image sent to the model. When static, a fallback frame is provided approximately every 20 seconds. Text spoken by the model via 'say' is displayed in a floating window and read aloud. A /seenow slash command toggles data collection.
Why It Matters When having AI watch your screen, it no longer captures full-screen screenshots every few seconds; static images consume almost no tokens, and scene switches are not missed.
Stars 576 (+47 / 24h)
Reasons
- 47 new stars in the last 24 hours (indicating attention level only)
- The README clearly specifies the token cost for each canvas size based on DeepSeek-V4's image chunking method
Usage
- Run
dsh plugin --profile web add dsh-see-now(use--profile desktopfor the desktop version), restart DSH after installation, and type/seenowin a session to start
Risks
- The plugin continuously captures screen or camera footage and sends it to the model; be mindful of privacy. The real-time glass-style floating window is CPU-intensive.
- Only applicable to DSH