Yes to every question you asked.
I'll clarify the current state of things here in a bunch of detail.
My personal view is that, over the long term, the blockchain is not an appropriate medium to store and communicate secret information. This is not just because those secrets are large (memos and key material) but also because post-quantum KEMs will consume most of the on-chain transaction size. It is a major scaling hazard.
The blockchain is for communicating state changes and agreeing on their ordering. That's all we should really need it for long term.
Tachyon is going to be decoupling the key structures and other parts of the on-chain *shielded* protocol (aka the parts that impact the actual private transfer of funds) from the *payment* protocol that is used to store and communicate the secrets used for those transfers. By doing so, we can use *many* different payment protocols and hopefully find one that scales and has good UX in the process.
Again, I want to use an out-of-band payment protocol that does not require users to rely on the blockchain for backup & restore or for detecting payments, since I don't think that scales. I don't have the time (this year) to ship a payment protocol and wallet UX that works nice according to my Glorious Future vision, but
@zkDragon has a team working on a post-quantum, on-chain payment protocol for Tachyon that uses PIR to fill in the gaps and approximate the existing UX without compromising privacy. And one of the things they are preserving is the ability to restore from the chain via your seedphrase.
People will be able to use Tachyon and the aforementioned post-quantum on-chain payment protocol by just running a full node, without depending on a third-party service. But the whole pitch for what we're doing is that (by using PIR and whatever other kind of indexing or proof generation services are needed) consumer wallets don't *need* to run a full node and yet can get the same privacy and custody properties they have with today's wallets.
The concerns upthread about there being services that wallets *need* to use in order to access their funds are a bit misplaced IMO. First, as I said, it will remain possible to run a full node and bypass that requirement. But consider that any blockchain which actually scales to a large number of people becomes out of reach for a massive percentage of its users to engage *directly* with, since these are inherently broadcast protocols... I am just following this to its logical conclusion.
This future may be directionally unavoidable. The (very attractive) UX of "restore from seedphrase from blockchain" is intuitive to me but not to many average people that try to use crypto, and so people might need to pay to obtain the UX (cloud backup? social recovery? etc.) they find intuitive. Blockchains can't securely and sustainably offer everyone on Earth these services.
But again, in the mean time, we're keeping things the way they are for Tachyon... restore from chain with your seedphrase.