Answers are grounded in this paper and cite their sources.
MeshStack is the distribution layer that lets THOX devices pool themselves into one private AI system. This report reconciles two internal design lineages into a single contract, capacity-priced pipeline sharding, grant-gated KV transfer, and lease-fenced device roles, and evaluates it with classed evidence: measured on a general host, calculated from public product floors, and simulated at protocol level. Measured results include lossless KV compression at 1.51x (zstd-3) with bit-exact round trip and HMAC chunk keying at 9.67 microseconds per 64-token chunk. The admission study's central result is negative and reported plainly: at the public device floors an Air+Mini+Clip trio (12 GB usable) cannot place a 27.8B Q4 model (14.6 GiB of weights), so generation-1 MeshStack is device-plus-hub for that model class. The paper claims no device or engine performance; every number is generated from the included experiment outputs.
No takes yet. Share an insight, caveat, or question.
Ross et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: