Consider adding pgvector extension

[ccjernigan]

pgvector enables vector similarity search which is useful for AI applications. Support exists in Azure, AWS, and Supabase for this extension, indicating it is growing in popularity.

I had originally hoped to use Upstash Redis on Fly.io for a similarity search, but Upstash Redis does not support this functionality yet. (Redis has the RediSearch extension for vector similarity search).

Pgvector support provides another way to achieve this.

[smorimoto]

That's not too hard to add on the end-user side. Shall I post an example Dockerfile? 🙂

[davissp14]

I'm not totally opposed to adding additional extensions to the base image, but we need to be a little careful about the extensions we add. We've had issues in the past with certain extensions that blocked our ability to release new major versions of Postgres. Some extensions are a bit more involved and can take the maintainers a bit more time to adapt to the changes that come with new PG versions.

We can always rip it out the extension in the event it's not compatible, but that ends up creating not-so-fun documentation issues

Anyways, we'll have to look a bit closer at this extension and determine how likely we'll run into the problem mentioned above.

[smorimoto]

I agree. That's why I think it makes the most sense to create a custom Dockerfile on the end-user side, build and deploy it as pg machine.

[tyrauber]

Hi @smorimoto, I'd love to see an example Dockerfile! It would be super helpful to know how to customize the pg instance without having to fork the repo.

Edit: Actually, this appears to the trick right here.

[lilouartz]

Has anyone actually managed to get it to work?