papers/hanzo-gpu-inference/sections/08-integration-with-hanzo-infrastructure.tex

\section{Integration with Hanzo Infrastructure}

\label{sec:integration}
% ============================================================================

The GPU inference system integrates with the broader Hanzo platform:
\begin{itemize}[leftmargin=1.1em]
    \item \textbf{Hanzo ACI}~\cite{hanzoaci}: GPU operators register as ACI
    compute nodes, staking HANZO tokens proportional to declared GPU capacity.
    \item \textbf{Hanzo Gateway}~\cite{hanzogateway}: All inference requests
    enter through the gateway, which handles authentication, rate limiting, and
    org-scoped routing.
    \item \textbf{Hanzo IAM}~\cite{hanzoiam}: Tenant isolation ensures that
    KV-cache, model state, and metrics are never shared across organizations.
    \item \textbf{Hanzo Edge Inference}~\cite{hanzoedge}: For latency-critical
    deployments, a subset of models are served at edge locations with the same
    scheduling framework.
    \item \textbf{Candle Runtime}~\cite{hanzocandle}: Rust-native inference
    kernels compiled per GPU architecture, eliminating Python overhead.
\end{itemize}

% ============================================================================