how it works
In the data path. One rack. Zero hops.
Ares does the work inline. Argus inspects, filters, routes. The model sees clean data.
stack swap
Before
cpu cluster path
agents / clients
cpu pre-processing tier (n+1 nodes)
bolted on, scales with load
model / inference
storage / logs
tail latency: unpredictable
After
nol8 inline
agents / clients
nol8 data plane (1 rack)
inspect · filter · route · log — in flight
model / inference
storage / logs
tail latency: flat
data in flight
01
arrive
event hits the wire
02
inspect
parsed at line rate
03
filter
junk and policy drops
04
redact
pii masked in flight
05
route
to model or store
06
deliver
clean → inference
live paththroughput 3.2M evt/sp99 1.4 msdrops 0.4%[ measured · single Ares rack ]
// content pending
FPGA architecture, throughput characterization, and policy DSL reference will land here. Gated until approved.