Grounded video guides · Media
Watchless
Long tutorials and courses are worth keeping, not re-watching, but an AI summary doesn't tell you which parts the video actually said and which the model made up. Watchless ties every step to the moment it came from and checks it. Paste a link and say what you want — a detailed walkthrough, a checklist, just one topic — and Gemini watches the video straight from YouTube, returning steps with a timestamp and the speaker's exact words. Each quote is fuzzy-matched against the captions within 90 seconds of its timestamp: a match marks the step ✓ and snaps it to the exact second; a miss is flagged ⚠ rather than hidden. ffmpeg then pulls the clearest 1080p frame of the UI, slide or code at each moment without downloading the video. A 3-hour course becomes a guide in about 4 minutes.
Highlights
- Every step carries a timestamp and the speaker's exact words, fuzzy-matched against the captions — ✓ found, ⚠ flagged, never hidden
- Screenshots of UI, slides and code pulled from YouTube's 1080p stream at each moment, without downloading the video; talking heads skipped
- Ask the video: answers come only from the video, cite the moments they rest on, and say “not covered” instead of guessing
- Guides shaped by your prompt — a detailed walkthrough, a checklist, or just one topic — kept in a local library
- Handles videos from a few minutes to 3+ hours; long videos read the transcript first and only look closely where it matters
- Export as Markdown with images, or save as PDF
Tech stack
- Python
- FastAPI
- Gemini
- yt-dlp
- ffmpeg
- pytest