(A question, not an issue) Do initial indexing in a different machine?

[rdiaz02]

Hi, great plugin! I am running the initial indexing on my collection, and the ETA is over 50 hours (I am using "Full paper" indexing because I very interested in finding specific passages). This is on a laptop, and that would preclude me from, say, restarting Zotero (or rebooting the laptop).

Since the embeddings are saved locally in zotseek.sqlite, I am thinking of installing Zotero on a server (with Zotero storage shared) and running the indexing there. I clicked on "Automatically index new item", and the two Zotero instances would be synced (the zotero.sqlite as per the standard Zotero mechanism, and I'd share the PDFs via Syncthing or WebDAV[1]); thus, I could continue using Zotero on the laptop, restart Zotero if needed, etc, and once the indexing is done in the server, copy zotseek.sqlite.

This sounds to me like an obviously good idea ( 😎 ) but I wonder if I am missing something, or if I could break something.

[1] Full details: all the PDFs are already on the server, copied there using Syncthing ---a relict of an old syncing mechanism I had. If I install Zotero on the server, I could just enable WebDAV, which is how I share PDFs with my tablet. But I think these details are irrelevant: what matters is that ZotSeek have access to the PDF.

[introfini]

Hi @rdiaz02, thanks for the kind words! This is a clever idea, and it can work — but there's an important technical detail to be aware of.

The gotcha: numeric item IDs

ZotSeek's database (zotseek.sqlite) uses Zotero's internal numeric item_id as its primary key for embedding lookups. These numeric IDs are local to each Zotero profile — they're auto-incremented integers, not part of Zotero's sync mechanism. So if you install a fresh Zotero on your server and sync your library, the same paper could get a different numeric ID on each machine, and the copied zotseek.sqlite would point to the wrong items (or no items at all).

Workaround that should work

Instead of syncing via the standard Zotero mechanism, copy your entire Zotero data directory to the server. This preserves the exact same numeric item IDs. Then:

1. Copy your Zotero data directory to the server
2. Install Zotero + ZotSeek on the server
3. Run the full indexing there
4. Copy just zotseek.sqlite back to your laptop's Zotero data directory (it lives at <Zotero data dir>/zotseek.sqlite)

Since the data directory is a direct copy, all the numeric IDs will match and the database will work as expected.

Important caveat: Don't add/delete items on the laptop while the server is indexing — any items added on the laptop after the copy won't exist in the server's database, and items deleted on the laptop may still have stale embeddings. After copying zotseek.sqlite back, new items will get indexed automatically (if "Automatically index new items" is enabled), and ZotSeek handles deleted items gracefully.

Which Zotero version?

One more thing — if you're on Zotero 7, the ~50h ETA is expected because its JavaScript engine (Firefox 115) has much slower WASM performance (~8-10x slower than Zotero 8). If upgrading to Zotero 8 is an option, you'd see a very significant speedup. You could also try Abstract Only mode first for a faster initial index, then re-index with Full Paper later.

Future improvement

The database does store the portable item_key (Zotero's universal string key that syncs across libraries) alongside the numeric ID, so making the database truly portable between synced Zotero instances is something I could add in a future version. I'll keep this in mind.

Hope this helps — let me know if you run into any issues!

[rdiaz02]

Thank you very much for your detailed response!!!

I am running Zotero 8.0.3 (under Linux), but I have about 9000 items and 32 GB worth of PDFs.
The "Important caveat" you mention for the procedure, though, makes it not viable for me, right now, as (work demands) I cannot not add/highlight/etc items in Zotero for the next 24 hours or longer. In general, not adding/modifying items for over 24 hours is not viable as of now, even on weekends. So I think the only option for me is to do the indexing in the laptop.

I've started by using the "Abstract only" (done in a few minutes) and then will re-index with Full paper.

I am really sorry for the noise and for wasting your time.

P.S. For those thinking about this, there is an additional issue to consider, unrelated to ZotSeek: if one runs Zotero in a headless server, one needs to figure out how to keep Zotero running there after the ssh connection disconnects. So one should look into things like Xpra or alternatives (which might add some complications).