Platform architecture and controlled AI
How SXP.ai delivers fine-grained AI control at every layer: orchestration, intent routing, micro-agent scoping, model flexibility, deployment, and the governance that keeps AI safe and accurate by design.
How the platform controls AI at every layer
SXP.ai separates orchestration, intent routing, micro-agent capability, content scoping, model selection, and institutional systems into distinct governed layers - so AI only ever does what it's been explicitly permitted to do.
Reference architecture
What this enables
- ✓ A single assistant experience that routes to purpose-built micro-agents
- ✓ Capability added safely - without widening what AI can access or do
- ✓ Campus systems connected through governed micro-agents
- ✓ Model choices that evolve over time without re-architecting
- ✓ Content scoped per assistant and audience
Intent detection & routing
End users see a single assistant. Behind the scenes, the platform detects intent and context, then routes to the right micro-agent - so every response is scoped, relevant, and aligned to approved institutional content.
One assistant, many capabilities
The assistant acts as an intelligent routing layer. It selects the right micro-agent based on what the user needs, who they are, and what channel they're on.
- ✓ Routes by intent, audience, channel, and policy
- ✓ Each audience gets a tailored experience
- ✓ Every response comes from a scoped capability
Scoped content and responses
Each assistant is grounded in approved institutional content. Content sources are scoped per assistant and audience - so a student-facing assistant doesn't surface internal staff policies.
- ✓ Content scoped per assistant and audience
- ✓ Grounded in institutional websites, documents, and knowledge bases
- ✓ No cross-contamination between audiences
Micro-agents for controlled AI
Micro-agents are how SXP.ai delivers fine-grained control. Each one performs a defined function with its own scope, permissions, and rules - so capability grows without widening what AI can do, say, or access.
Guardrails and safe usage
Micro-agents operate within defined scope, with multiple layers of control - from architectural constraints to conversation-level guardrails - supporting safe operation in sensitive and high-risk contexts.
- ✓ Scoped access & permissions
- ✓ Business rules and workflow controls
- ✓ Logging and auditability
- ✓ Human follow-up and escalation pathways
Pre-built micro-agent library
Pre-built micro-agents for common higher education use cases - enabling rapid deployment with minimal configuration.
- ✓ Wellbeing support & structured journeys
- ✓ Retention patterns & proactive engagement
- ✓ Teaching & learning support
- ✓ Q&A over websites, documents, and knowledge bases
- ✓ Service automation
Custom micro-agents
Additional micro-agents can be configured or delivered as separately scoped professional services - maintaining clear governance, cost control, and delivery assurance.
- ✓ Separately scoped delivery and acceptance criteria
- ✓ Explicit permissions and data access design
- ✓ Operates under the same oversight and logging model
Deployment and integration
SXP.ai fits into your existing digital estate. Assistants authenticate through your identity provider, deploy wherever your users already are, and connect to campus systems through governed micro-agents.
Deploy anywhere
Assistants embed into your existing website, student portal, or VLE - or run as a standalone full-screen experience.
- ✓ campusM, myday, SharePoint, and institutional portals
- ✓ VLE/LMS integration via LTI where applicable
- ✓ Mobile-first responsive design
Identity & authentication
Integrates with your existing identity provider so users authenticate once. Assistants tailor responses based on role, cohort, or entitlement.
- ✓ SAML 2.0 and OpenID Connect (OIDC)
- ✓ Role-based access and audience segmentation
- ✓ Supports authenticated and anonymous modes
Campus system connectors
Micro-agents connect to institutional systems to retrieve information and trigger workflows - without exposing those systems directly to AI.
- ✓ Service management platforms (e.g. TOPdesk, ServiceNow)
- ✓ Student information systems and CRMs
- ✓ Workflow triggers and case handling
Content & knowledge management
Assistants are grounded in your approved content. Knowledge sources are managed centrally and kept up to date without retraining models.
- ✓ Websites, PDFs, knowledge bases, and documents
- ✓ Managed ingestion with version control
- ✓ Content scoped per assistant and audience
Model flexibility without lock-in
The platform is model-agnostic by design. Micro-agents can operate across Amazon Bedrock, Claude, ChatGPT, or institution-provided models - allowing you to choose, mix and change models as the landscape evolves.
No vendor lock-in
Choose models that align with policy, risk appetite, data residency, and cost - and change those choices over time as governance evolves.
- ✓ Bedrock-first deployments available
- ✓ External LLM support where approved
- ✓ Ability to adopt new models safely
Policy-driven selection
Model choice is implemented within the governed architecture - supporting central IT oversight and reducing fragmentation.
- ✓ Platform-level controls
- ✓ Consistent governance enforcement
- ✓ Repeatable implementation patterns
Governance, compliance, and responsible AI
EU AI Act governance, risk management, and oversight principles for all classifications - including high-risk - are embedded directly into the platform. This removes the need for universities to design bespoke AI governance frameworks for each new use case.
Built-in oversight
Oversight, policy enforcement, logging, and auditability are embedded at every layer - built in, not bolted on.
- ✓ Guardrails and logging by design
- ✓ Defined escalation pathways to human support
- ✓ Visibility into micro-agent behaviour and decisions
- ✓ EU AI Act high-risk classification support built in
Security and certification
Hosted on AWS and delivered under certified information security and privacy management systems.
- ✓ ISO/IEC 27001 certified
- ✓ ISO/IEC 27701 certified
- ✓ GDPR compliant
- ✓ EU AI Act compliant
Live in days
SXP.ai is delivered through packaged services - not left to your team to configure and deploy. Universities get production-ready AI assistants without needing internal AI engineering capacity.
What SXP.ai delivers
Pre-built micro-agents, assistant configuration, content ingestion, deployment, and ongoing platform operation - delivered as a managed service.
- ✓ Pre-built HE capabilities ready to deploy
- ✓ Assistant branding, scoping, and configuration
- ✓ Content ingestion and knowledge base setup
- ✓ Deployment into your channels and platforms
What your team provides
Access to content sources, stakeholder input on scope and audiences, and sign-off on governance and deployment - everything else is handled by SXP.ai.
- ✓ Content and knowledge source access
- ✓ Audience and scope requirements
- ✓ SSO / identity provider configuration
- ✓ Governance sign-off
See how it works
A walkthrough tailored to your institution - whether you want a quick overview or a deep dive into platform architecture, governance, and deployment.