Meeting notes export
gnt prebrain --meeting-notes reads meeting-transcript exports from Otter, Fireflies, or Fathom.
Point it at a directory or a single file and it auto-detects each file's shape, no per-tool flag
needed, no gnt server ever in the read path.
Otter
Open a conversation, use the export option in its menu, and choose a format. Every plan can export TXT; paid plans add DOCX, PDF, and SRT. Either TXT or SRT works here, no need to pick one over the other for this to read it.
Fireflies
From a meeting's notepad, use the download option and choose a format: DOCX, PDF, SRT, CSV, JSON, or MD. SRT or MD both work here; turn on the timestamp and speaker-label toggles if the export screen offers them, since a transcript reads better with both.
Fathom
Fathom has no file export at all. From a recorded call, the Copy Transcript button above the
transcript copies it to your clipboard, and that's the only built-in path. Paste it into a plain text
file and save it with a .txt extension, then point --meeting-notes at the folder you're saving
those into.
Point it at your exports
A directory of exported files, or a single file, both work. .vtt, .srt, .txt, and .md files
are read; which one a given file actually is gets sniffed from its own content, not trusted from its
extension, so a mislabeled file still parses correctly.
Code
What it reads, and what it never does
A cue-timestamped file (SRT, or VTT if a tool you bridge through produces one) has its cues merged back into full spoken turns before anything else happens. Caption exports slice a single sentence across several short cues, and reading each cue as its own turn would fragment ordinary dialogue and misread where one person stopped talking and another started. A plain-text file's speaker and timestamp markup is read the same way, whichever of Otter's, Fireflies', or Fathom's line shapes it uses.
Same as every other prebrain walker: everything above happens on your device before the privacy gate runs, and only masked text ever reaches extraction.