Vectorize feedback
Hey there! Vectorize has been out a month now. We'd love your feedback if you have some that you haven't shared yet. What's good, what's bad, what could be better, we take it all.13 Replies
If you answer, please do in this thread 
I've only used it for https://superspeare.cloudflare.community, but we didn't have any problems with it during that project. I really liked how easy it was to make an index compatible with the models on workers AI. The only issue we encountered was that you couldn't get the metadata without also fetching the vectors, but I believe this has since been added.
@zegevlier very nice tool 🙂 https://superspeare.cloudflare.community/?query=awesome
The biggest blocker for me is not being able to sort and filter vectors by metadata.
Unknown User•16mo ago
Message Not Public
Sign In & Join Server To View
does it run locally in development? or only on a deployed worker?
The key here is scale with simplicity. I love that we dont have to think about replication strategies and it all just works. I think that should actually be the biggest part of the message you sent out to folks. Super impressed thus far - congrats team!
No local dev yet
My biggest piece of feedback would be for CF to launch all new features in a way that calls out to remote services in local dev (like how preview KVs work). I don’t wanna have to switch to remote dev or hook up my own services just to adopt a new feature.
Would love to be able to try it out for our use case but the current TopK limit is way too small
I’d also like pagination on the top K query—if a result in the top K isn’t suitable, it would nice to be able to fetch the K+1st result without computing the original K
We're working on it 🙂
it does not run locally (no simulator); it uses edge preview for wrangler dev
@dook.dook what value of topK limit would fit your use case?
Oh and one more—just easier ways to manage vectors in bulk, things like a
list endpoint with similar filtering to the query endpointWe're using it for semantic search so ideally as high as possible in prod. I think our max results top out at like ~300 at the moment. But for testing like 50 - 100 would be great. Cheers!