File Save Not Working In Gemini

[markran]

Platform: Windows 11 x64
Firefox: 152.01
LLM: Gemini Webchat (free acct / logged in)
Continuum: v1.1.1 ('Always Allow' active, all permissions granted)

I've tried saving files from two different sessions:

Session 1 (2 round session, no media): 303KB ZIP file which 7-Zip File Manager shows as having 2,693 folders and 0 files. The structure is a top folder labeled transcript.md containing all the other folders named "1" - "2692"

Session 2 (26 round session, some images): 5.61MB ZIP file which 7-Zip File Manager shows as having 50,065 folders and 0 files. The structure is a top folder labeled transcript.md containing all the other folders named "1" - ""50,064"

In both sessions the saved file structure and size remains the same whether I saved it as a PDF or MD (top folder named "transcript.md").

I did a Factory Reset of the add-on after a few attempts at saving Session 1 and had identical behavior again with Session 1 before proceeding to trying Session 2.

Note: The message count shown for Session 2 was initially 20 on my first 3 save attempts. It occurred to me the session was probably longer, so I scrolled the Gemini chat window back up to the top of the session to check. This was far enough it caused Gemini to do a page reload. I then repeated "Capture This Session" twice and both new captures of Session 2 now showed 26 messages. File saves from Session 2, Captures 2 and 3 showed 6.96MB and 62,123 folders (still 0 files, no change whether PDF or MD). This may be a separate issue of the browser only giving you a partial session if it's not all currently in memory. My scrolling the session back up and before doing capturing again apparently forced a more full reload. I've previously used the add-on "AI Exporter" and it has saved very long sessions without scrolling to the top.

Feature suggestion: It would be nice to allow setting the filename before saving the file.

[mofe-stack]

Thanks for the detailed report! You basically diagnosed both of these yourself, and both are fixed in v1.2.0 which is live now and has a few more features and bug fixes you can check out.

The "thousands of empty folders" ZIP was a Firefox-specific bug in how the archive got written. Continuum builds the ZIP in-page with fflate, and on Firefox the TextEncoder output (your transcript bytes) came back as a Uint8Array from a different JavaScript realm than the one fflate was running in. fflate didn't recognize it as file content, so it fell back to treating each entry as a directory path which is exactly why you saw transcript.md as a folder full of nested "1", "2", … folders with zero real files, scaling 1:1 with the message count. The fix was to re-wrap the encoded bytes in a fresh same-realm Uint8Array before handing them to fflate, so it sees real file data again. You'll now get a proper transcript.md plus images/ and files/ folders with the actual content inside like Chrome. In addition I fixed the bug where the extension on Firefox wasn't capturing images and files.

The partial Gemini session is the other thing you pinned down perfectly in your note: Gemini virtualizes its chat, so it only keeps the on-screen turns mounted in the DOM and unloads the rest meaning a capture from the bottom only saw whatever happened to be loaded (hence the 20 to 26 jump after you scrolled up and forced a reload). I reworked capture to do a hardened full-history pass first which takes a couple of seconds. It scrolls the entire conversation while watching the scroll height, scroll position, and turn count until they hold steady, so every turn is mounted before anything is read with no manual scrolling needed, wherever you start. Longer term I'd like to capture Gemini and Perplexity straight from their network/RPC responses which I think what that "AI Exporter" extension you (pulling the data directly, no scrolling at all), but the hardened scroll path is reliable and is what's shipping for now.

And good call on the filename suggestion letting you set the name before saving is on the list. Thank you again for this detailed report which helped me fix these issues 🙏