为 AI 而构建的系统基础设施。
Purpose-builtAI infrastructure.
为 AI 而构建的系统基础设施。
Purpose-builtAI infrastructure.
从高带宽公网入口、分布式执行与状态，到统一 Intelligence Plane 与模型系统协同。一个纵向设计的 AI Execution Fabric。
From high-bandwidth edge and distributed execution to a unified intelligence plane and model-system co-design — one vertically engineered AI execution fabric.
公网入口保持简单，业务权威留在核心。
Keep the edge small.Keep authority inside.
公网入口保持简单，权威留在核心。
Keep the edge small.Authority stays inside.
Edge 只负责公网 TLS、连接接入与有序字节流转发。配置、策略与持久业务状态继续由核心系统掌握。
The edge handles public TLS, connection ingress and ordered byte forwarding. Configuration, policy and persistent product authority stay in the core.
按职责扩展，而不是扩展整个集群。
Scale each role,not the whole cluster.
按职责扩展，而不是整个集群。
Scale each role,not the wholecluster.
Controller、Model Gateway、Node、Data Gateway 与 Storage 各自形成独立扩展域，共享一个事务化权威状态。
Controllers, model gateways, nodes, data gateways and storage scale independently around one transactional source of product authority.
更少的层，更明确的边界。
Fewer layers.Explicit boundaries.
更少的层，更明确的边界。
Fewer layers.Explicit boundaries.
YaoBox 直接围绕 AI 执行组织 Process、Network、Filesystem 与 Resources，不引入与产品无关的通用编排抽象。
YaoBox organizes process, network, filesystem and resources directly around AI execution, without carrying unrelated orchestration abstractions.
统一 Intelligence Plane，连接多种模型后端。
One intelligence plane.Many model backends.
一个统一的Intelligence Plane，连接多种模型后端。
One intelligence plane.Many modelbackends.
请求 staging、路由、Provider 执行、流式解析与标准化被收敛到统一网关；外部、自托管与自研模型共享同一调用表面。
Staging, routing, provider execution, stream parsing and normalization converge in one gateway across external, self-hosted and internal models.
故障显式发生，恢复确定完成。
Failure is explicit.Recovery is deterministic.
故障显式发生，恢复确定完成。
Failure is explicit.Recovery isdeterministic.
先持久化 authority，再允许 external effect；结果不确定时通过 readback 与 reconcile 恢复，而不是盲目重放。
Authority is persisted before external effects. Unknown outcomes recover through readback and reconciliation — never blind replay.
模型与系统，是同一个工程问题。
Models and systemsare one engineering problem.
模型与系统，是同一个工程问题。
Models and systemsare one engineeringproblem.
Architecture、Numerics、Communication、Runtime 与 Serving 被放在同一工程表面协同设计，从训练数值一直贯穿最终执行。
Architecture, numerics, communication, runtime and serving are co-designed on one engineering surface from training precision to execution.
一个技术底座，承载不同产品形态。
One foundation.Multiple products.
一个技术底座，承载不同产品形态。
One foundation.Multiple products.
Infrastructure、Distributed Runtime、Intelligence 与 Model Research 向上连接支付、分发与真实产品，把系统能力带入商业环境。
Infrastructure, distributed runtime, intelligence and model research connect upward into commerce, distribution and real products.
