News for hackers. Stuff that matters.

Ask HN: In-Browser Vector DB?

I'm looking to develop a retriever-augmented LLM Q&A app that will call out to an API for the LLM (this will be the only allowed type of API call [other than a fileserver], it's internal and no data can leave the firewall). I plan to ship the chunked, preprocessed document store as embeddings with the app (~100MB, can be read as files from a fileserver).

Is there any way to ship a vector DB implementation in-browser?

Other research:

* I know that Chroma can use DuckDB under the hood, and DuckDB has a WASM port, but Chroma itself doesn't?

* Weaviate and Qdrant do not have a WASM target?

( 6 comments | 2 points | on HN )
Re:Ask HN: In-Browser Vector DB? by dhruv_anand on Monday June 05, 2023 @01:00PM (#36199314)
Re:Ask HN: In-Browser Vector DB? by ukuina on Monday June 05, 2023 @03:59PM (#36202659)
Thank you for the pointer. This still requires a backend server running Chroma, correct?
Re:Ask HN: In-Browser Vector DB? by andre-z on Monday June 05, 2023 @12:57PM (#36199270)
Hi, Qdrant team here. Could you please explain why it has to be in-browser and not just start a docker locally?
Re:Ask HN: In-Browser Vector DB? by ukuina on Monday June 05, 2023 @03:57PM (#36202630)
Thank you for responding here.

For this enterprise use-case, I am looking to ship a web app that will run off of IT-managed fileservers and IT-managed LLM APIs, on user desktops and laptops without privileged access to install software.

Re:Ask HN: In-Browser Vector DB? by andre-z on Tuesday June 06, 2023 @11:17AM (#36214162)
If the dataset isn't extensive, you might be able just to use an ANN library like FAISS.
Re:Ask HN: In-Browser Vector DB? by ukuina on Tuesday June 06, 2023 @10:00PM (#36221889)
That's a great idea, I will look at FAISS for now. Thank you.