Reference Architectures
Blueprints we would actually build
Each one is a working design document, not a diagram: trust zones, connection matrices, the decisions that were closed, and the ones still open. Take them, argue with them, ship them.
Governed agent platform
Enterprise AI Agent Frameworkv1.0
A self-hosted framework where agents may reason autonomously while access to models, enterprise systems, code execution, and irreversible actions stays under infrastructure control. An architecture brief for decision makers, then the technical reference behind it.
- Controls
- Models, tools, code, approvals
- Harness
- Deep Agents on LangGraph
- Principle
- Autonomy in the agent, authority in infra
Enterprise model infrastructure
Inference Fabricv1.0
A dedicated, governed serving estate for operating open-weight models as internal infrastructure — one controlled ingress, inference-aware routing, and capacity that belongs to a workload rather than to whoever asked first.
- Contract
- OpenAI-compatible model APIs
- Routing
- Cache-, load-, and workload-aware
- Rule
- One ingress; no direct model endpoints
GPU inference fabric
Multi-Cluster Inference Fabricv0.1
A GPU inference fabric that survives the loss of a rack, a cluster, or a region — active-active DGX-class capacity behind one routing control plane, with parallelism strategy and failover behavior specified precisely enough to build from.
- Clusters
- 2+, active-active
- Hardware
- DGX-class · 400G+ IB or RoCE
- Constraint
- No parallel group crosses a cluster