I am doing a metatrancriptomic analysis, now I am analyzing the sequences corresponding to a parasite that live inside a host. I have millions of reads to the host but fewer reads of the parasite.

My library size of parasite by replicate is around 193,000. I have seen that "As a rule of thumb, we require that a gene have a count of at least 10–15", given my low deep I can filter by cpm>1. What can I do, use a higher cpm value?

Yes, you will need to use a higher cpm cutoff if a sizeable group of your samples have low sequencing depth. You can use the provided function, filterByExpr, which will choose an appropriate cpm cutoff for you automatically.

The sentence you quote is from here:

https://f1000research.com/articles/5-1438

There is an updated version of the same article here:

RnaSeqGeneEdgeRQL vignette

and you will see that the filterByExpr function is now used.