Metis AERIX: Governed Hybrid AI
One secure gateway for policy routing, model selection and full audit visibility.
Overview
Metis AERIX extends the Metis ARX private AI core with cloud AI capability, under governance. Policy routing keeps customer data local. Public-safe work can use approved cloud models. Every routing decision is recorded in the audit trail.
For regulated organisations, this addresses a common tension: they want the speed and breadth of cloud AI, but their data and their regulators do not allow sensitive content to leave a controlled environment. AERIX lets them use both sides, with the organisation setting the rules.
The problem AERIX addresses
Staff already want AI in their daily work, and many paste sensitive text into public tools because no approved option exists. Boards remain accountable for every automated decision, and under the principles of both HKMA and MAS that accountability cannot be outsourced to a vendor. Organisations therefore need to prove control over where data goes, which model handles it and who approved the outcome.
How AERIX works
AERIX is a single gateway between an organisation's applications and every AI model it uses. Apps, mobile channels, users, analytics and integrations connect once, to AERIX, and the gateway selects the right model for each request.
It applies four controls:
Model selection. Each request goes to the model best suited to the task.
Automatic switching. Requests move between cloud and private models as conditions require.
Policy routing. Rules decide which content may reach the cloud and which must stay private.
Fallback and continuity. If a model is unavailable, service continues through a defined alternative.
Two sides of AI, one governed model
| Cloud AI | Private AI (Metis ARX) | |
|---|---|---|
| Strength | Scale, speed, latest capabilities | Privacy, control, custom knowledge |
| Typical content | General knowledge, language fluency, public research | Raw customer data, contracts, internal documents |
| Models | Cloud LLMs and specialist models | Private model, private LLM, private data |
| Delivered by | Public hyperscale cloud | On-premises appliance or private cloud |
A secure hybrid-cloud bridge connects the two sides. On the private side sit air-gapped data control, deterministic regulatory rules, and PII sovereignty with local audit. Sensitive work is handed over to this side deliberately and traceably.
Four benefits
Multi-LLM freedom. Organisations are not locked into one model or vendor.
Data protection. Customer data stays local; only approved, public-safe work reaches the cloud.
Always available. Fallback routes keep services running.
Full visibility. Each decision is logged, so compliance and risk teams can see what was routed where, and why.
Built for regulated environments
The Metis platform is certified to ISO/IEC 27001 (information security) and ISO/IEC 42001 (AI management systems). Its controls map to the expectations of HKMA and MAS, including:
Decision traceability and audit trail
Human-in-the-loop approval
PII masking, tokenisation and data residency
AI firewall and model governance
Prompt-injection protection and data-leakage prevention
A deterministic rules engine for agentic workflows
This describes how Metis controls map to these frameworks. It is not regulatory endorsement.
What it looks like in practice
A relationship manager asks for a summary of a public market report and a draft reply to a client email that quotes the client's account details. AERIX routes the first request to a cloud model, because the content is public. It keeps the second on the private ARX core, because it contains client data. Both routing decisions are logged, and a person reviews the reply before it is sent. The same pattern applies to Metis Ally in Outlook, the Messenger Agent on business messaging channels and Tender Intelligence: generic tasks may use cloud AI, and anything with client data stays on private AI.
Deployment options
On-premises or air-gapped. Maximum sovereignty, including a desktop-class reference workstation with no server room.
Private cloud. Managed, high-availability infrastructure for fast scaling.
Hybrid through AERIX. Sensitive processing stays local, and public-safe work uses cloud models, routed by policy.
A typical implementation takes 16 weeks from setup to go-live, with training and hypercare included. Customers can start with a pilot that pairs Metis Ally and Tender Intelligence on an Essential-tier node, then add OCR and RAG workflows, the Certificate Agent and more departments. AERIX policy routing, audit trail and FinOps cost visibility apply from day one.
Explore the platform
To arrange a demo or start a pilot, contact enquiry@metis.info.
About Metis
Metis is a private and hybrid AI platform for banks and other regulated enterprises.
Explore the platformDefine your next step with Metis.
Discuss your operational priorities, deployment requirements and evaluation criteria with our team.
