How do you deal with security and access control across postgres and pinecone?

We use Cloudflare Workers for our API and just handle auth calls by checking the JWTs with Supabase and caching it. So we already had the necessary auth setup to do this.

For basic CRUD we use the Supabase endpoints directly but none of that involves querying a vector db :P

Another option is to cold email investors in case you are open to working at a start up.

So much hate :( I agree the repo needs some clean up to be useful. Nevertheless, good collection!

Key Insights:

1. Many database back up their data in a sqlite database. Some even push vectors into sqlite, but others store vectors in their own format.

2. Qdrant has higher client connection and index initialization time that can shadow its benefit on fast and accurate vector search.

This article contains a lot of inaccuracies.

Based on your statements, like

> Qdrant stores both the vectors and the metadata in a sqlite database.

It looks like you have benchmarked local mode of qdrant. It doesn't even use vector indexes and is not designed for any kind of production usage.

For anyone reading this article, I urge you to do your own benchmarks and not rely on claims that do not have open source code attached to them to replicate the results

Hi Andrey. Thanks for your feedback. We should have better emphasized that we are benchmarking Qdrant in local mode. We have updated the post to clarify that Qdrant is being evaluated in local mode. We plan to next evaluate the server mode.

We went with the local mode as several Python AI apps are using Qdrant in that mode based on the suggestion here: https://qdrant.tech/documentation/quick-start/.

We also believe in open-sourced benchmark code. Please find the code here: https://github.com/jiashenC/vectordb-benchmark-and-optimize/....

Are you using RAG to synthesize responses?

Made it easier to document code. You can send the function as context, and it will generate a well-formatted docstring that accomplishes 90% of the task!

I don't believe there is a solution otherwise. All you can do is cross join.

I appreciate your view on using LLMs in production. However, I think the ability to use ML/AI directly within databases is a significant step forward. Most production systems already rely on predictions, so the idea of using them without leaving the database is both interesting and valuable.

Awesome work! What is the preferred way of adding embeddings in sqlite table?

Impressive! Quick question - is it possible to generate sql for a slight variant of sql? My project augments standard sql with a few new constructs.

It probably could but it would require adjusting the prompt. You’d have to override the generate_prompt function and tell it that you’re using a variant of SQL and describe the differences.

Is there a token limit for the train step? Like length of documentation, number of example SQL queries?

There isn't technically a limit on the storage side but it's generally better if you keep documentation to a manageable length.

You call vn.train(sql=...) on each individual SQL statement that you have.

What'll happen under the hood is the package will use the 10 most relevant SQL statement examples, 10 most relevant pieces of documentation, and 10 most relevant DDL statements.

If using 10 examples exceeds the (approximate) token limit for the model, it'll pare down to a smaller number that'll fit into the context limit.