If you can lock to a single-writer multi-reader model, SQLite is capable of handling very impressive loads.

This works especially well with horizontal scale apps where, for example, you can shard by client and want to have an isolated environment per-client.

Then you can do pg FDWs with a partitioned table and aggregate all of the SQLite dbs back together in a single view for super simple and fast cross-shard fan queries.

I think the hardest part for these sorts of projects is figuring out authorization.

For the two best-in-class (IMHO) examples of how to deal with this check directus and supabase. They have very different approaches to the problem, both very good.