What is the Proto-Proto-Indo-European?
The modern Moscow school asserts that PIE was predated by Proto-Eurasiatic. It is quite reconstructable at least in terms of vocabulary (examples include *apa for "father", *'aku and *wete for "water") and some other features (-s ending for the Genitive, *mi(n) for "I" and *ti for "thou", *ku and *io interrogative pronouns, ket(a) for "a pair"). Unlike PIE, numerals cannot be reconstructed for Proto-Eurasiatic which suggests that people did not know counting yet at the time.
The families, included in Eurasiatic are as follows:
It is further asserted that Proto-Eurasiatic is predated by Proto-Nostratic which includes
A large part of Nostratic vocabulary has been reconstructed, you can find a dictionary here: http://www.dspace.cam.ac.uk/bitstream/1810/196512/49/00ND_ALL.pdf
Further speculation is that Proto-Nostratic was predated by Borean, which includes Nostratic and Dene–Daic (which is subdivided into Dené–Caucasian and Austric).
It's sort of like asking a physicist "what's inside the smallest particles we've been able to observe?". By definition, we don't know. Proto-Indo-European is as far back as our current methods are able to reconstruct, so anything before that tends to be shrouded in speculation.
However, this doesn't mean it's unknowable—it just means it's the limit of our current evidence and current method. If someone comes up with a new and improved version of the comparative method, for example, they might be able to show some more conclusive evidence for "Proto-Eurasiatic" or "Proto-Nostratic" or the like. Or, if we find some evidence for a new related language, our current models could project even further back in time; that's what happened when Hittite was discovered.
The Eurasiatic hypothesis is very logical, since the Middle East have been a sort of spread sentral for the world population. But this language surely had numerals, at least up to 10, and probably far higher since counting is an obvious act of human intelligence and since the number of our fingers easily suggest the need for numbers up to 10. But at the earliest stages the numerals probably had a strong tendency to be shifted out with new ones.