correa_brian

For vector search we couldn't get my results to return meaningful entries. My preference would have been to use sqlite. Any others you recommend besides sqlite with the vss extension?

How so?

Haven't looked back since

For now we're using CLIP. We've also done testing with Siglip and Gemma for a full-blown vision model.

dont like the sound of reimplementing the whole UI components....

I love Gemma. I use it on LM studio and am working on getting it into Desktop Docs. Thats for the nomic link. I'll do some testing...

I've heard good things about Flutter! We briefly considered it but just opted for Tauri out of preference.

The move from Electron helped us remove "native libraries" in Node. Previously we were using Xenova transformers to instantiate the models. This forced us to use multiple package.jsons and added a lot of cognitive overhead to our development.

The move to rust freed us up to focus more on feature development than configs and setup. It was surprising because I thought learning rust would set us back much longer, but the trade-off was worth it.

Great resources, thanks. I'll look into the other packages and compare against our onnx runtime setup.

Nice. I'll try to test these out against our current implementation.

I did some prototyping with SQLite + VSS extension to handle embeddings and wasn't getting the same results as using the Redis search modules. My preference would be to run something that doesn't require a separate for the DB, but it still needs some tuning. Redis gets the job done but there are likely simpler solutions that could give us comparable performance that we'll discover.

Desktop Docs needs a GPU to work well, so we started with the Silicon chip Macs since they are ideal candidates to handle the workload of indexing an entire media library.

What do you think would be more in line with what you've seen in the past? There's definitely room to keep optimizing.

Yeah we didn't get a lot of traction on PH unfortunately and they have a lot of rules about relaunching. As far as users go - it ranges from prosumers to producers at media companies.

For sure. We've been prototyping embedding a vision model in desktop docs, but just needs more dev time to be stable. Went with CLIP for parity, but we are looking to upgrade soon. I tried Siglip and wasn't impressed. Do you know other open-source image embedding models you'd recommend?

That's it! Almost one year anniversary for Desktop Docs!

Thanks for the feedback! Will get a demo video and contact info up shortly.

De-duplicating images is on our roadmap. Shoot me your contact info at hello@desktopdocs.com. Would love to see if we can help.

Should we do a c++ rebuild

We considered it and decided all the wrangling wasn't worth it. Briefly also thought about adding in a python process to interact with our ML models, but figured it was worth streamlining this all into a single language/framework vs. creating frankenstein's monster.

Awesome! Linux version is on our roadmap and that use case is exactly why we started building this too. I can let you know when that version is ready

We haven't started testing for windows yet. Are you on Windows? Happy to let you know when we're releasing that version.

After our Electron experience, I wish I had moved out of my comfort zone (JS) sooner. Electron just requires a lot of optimization and you have to be really tight with your imports to avoid loading things you don't immediately need.

The smaller bundle size with Tauri and blazing speed are well worth the effort.

Thanks! If you're interested in that version, drop us a note: hello [at] desktopdocs dot com. We'll shoot you an update when it's ready!

Yeah there are definitely hang ups along the way. It's a big of wack-a-model but thankfully we've had an easier time developing with Rust than on Electron.

We aren't offering a trial right now, but to answer about docs --> we're testing an update to support searching text and will be releasing it soon. Right now you can search the contents of all your images and videos.

Thanks! Mentioned it on the other comment - but we're using the Ort crate in rust and bunlding onnxruntime with the app. Definitely considered Swift and I know it's gotten a lot better since I last used it, but cross-platform support was what got us to use Rust over Swift.

As far as porting over goes, we are much happier maintaining the new version.

We're using the onnxruntime bindgs with the Ort crate. Our biggest challenge was just bundling the onnxruntime into the app and making sure everything was signed, etc.

Thanks for checking it out. The goal is for Desktop Docs to be cross-platform. We've had a lot of requests for Windows support, so we chose Rust to set us up for an upcoming Windows version.

You have to charge for upgrades. It's not reasonable to pay once and expect lifetime updates. Subscriptions are annoying, but if you want ongoing support of course you have to pay.

This is cool. Can I bring my own vector database? I'm currently prototyping and want to experiment with different solutions. Would be ideal if I could avoid needing to implement a few different SDKs.