Check media and GitHub text
Transcribe video, audio, or YouTube, or fetch GitHub prose, then run detection.
Install ffmpeg and whisper-cli (whisper.cpp). The first media run needs
--download-model, or an interactive yes, to fetch the pinned ggml weights.
pangram video ./lecture.mp4 --max-billable-units 80 --download-model
pangram audio ./interview.wav --model large-v3-turbo-q5_0 --max-billable-units 40
pangram youtube 'https://www.youtube.com/watch?v=...' --max-billable-units 80youtube runs your own yt-dlp. Pangram CLI does not install it, and using
it is your YouTube-access decision.
GitHub shortcuts read GH_TOKEN or GITHUB_TOKEN:
pangram pr owner/repo#123 --max-billable-units 5
pangram issue https://github.com/owner/repo/issues/9 --comments --max-billable-units 10
pangram comments owner/repo#123 --max-billable-units 10Review comments send only their body, never the diff hunk. Output uses
origin: transcript or origin: github.
Paste a URL directly
A bare pull request, issue, or YouTube URL runs the matching shortcut. Put the shortcut's flags after the URL:
pangram https://github.com/owner/repo/pull/123/files --comments --max-billable-units 5
pangram https://github.com/owner/repo/issues/9 --max-billable-units 10
pangram 'https://youtu.be/...' --model large-v3 --max-billable-units 80--max-billable-units is still required. Any other bare URL fails with
unsupported_input instead of being analyzed as text. To analyze a URL as
text, use pangram detect 'https://example.com/post'.