I thought their relay cache thing was open sourced

Even if it's hidden from the user, you can always open devtools and switch to the Network | WS tab

The primal client doesn't connect direct to relays, it connects to a cache that's populated with unknown list of relays on the backend.

FBI relay confirmed

They are trying to make the relays fade away

Their client connects to a caching indexer service. AFAIK they have plans to open source it and allow you to connect to your own in their client.

Look here. It's open source now and they list which relays they are connecting too.

https://github.com/PrimalHQ/primal-caching-service

I believe I’ve heard someone say that primal is not technically a nostr client

Ser, it’s a nostr layer 2

Primal clients connect directly to your list of relays for all publishing actions. We haven't yet implemented the relay settings UI; that's coming very soon.

When it comes to reading, Primal clients connect to our caching service. Our plan is to enable users to connect to any caching instance, as well as offer the ability to disable caching altogether, in which case all reading would be done directly from the relays. We view the ability to work with a caching service to be an optional competency that some clients can choose to implement, *not* as a replacement for connecting directly to relays.

Note that the entire Primal software stack has been open sourced under the MIT license, so everyone is free to modify it and further experiment with these techniques.