System Architecture
Three layers. Governance, execution, and presentation. Independent and scalable.
Partner Platform
Business logic & data
Hub Brain
Governance & orchestration
Embedded Agent
AI execution layer
End Users
Governed AI interaction
Layer 1
Governance Layer - The Hub Brain
The central control plane that enforces policy across all AI operations. No agent executes, no knowledge is retrieved, and no action completes without passing through this layer.
- Multi-tenant data isolation and RBAC
- Capability manifests and approval chains
- Contract enforcement and usage metering
- Provider-agnostic LLM routing
- Full audit trail and decision tracing
Layer 2
Execution Layer - Partner System
The partner's own platform where business logic resides. The Hub Brain integrates with this layer through secure APIs, receiving context and returning governed AI responses without replacing existing infrastructure.
- JWT-secured session initialization
- Tenant context propagation
- Action callbacks and webhook handlers
- Data source connectors for RAG
Layer 3
Presentation Layer - Widget SDK
The user-facing AI interface embedded inside partner applications. Iframe-isolated for security, white-labeled for brand consistency, and governed by the platform for every interaction.
- Iframe sandbox with CSP headers
- PostMessage-driven communication
- Theme-aware, responsive rendering
- Session-scoped AI context