Deploy models for your customers
Deploy supported models on Aventail™ Link Server, with configuration managed through Aventail™ Control.
Deliver managed AI.
Across your infrastructure and theirs.
Give enterprise customers a managed AI service that reaches beyond your data centre. Aventail™ Server combines an installed inference runtime with deployment, application access and policy-based routing across your compute and the customer’s approved infrastructure.
Your own model execution. One managed system.
Aventail™ Link Server runs supported models on your servers. For supported models and hardware, it can replace a standalone serving deployment such as vLLM, bringing deployment, endpoints, policy-based routing and Aventail™ Control management together.
- CustomerWorkflows and applications
- Where the request startsAventail™ LinkLocal policy · own model execution
- One eligible destination is selected
- Customer infrastructure · Aventail™ Link
- Your servers · Aventail™ Link Server
- Approved external API
Aventail™ Control distributes agreed configuration ahead of time, outside the request path.
Install Aventail™ Link Server on your compute.
Run Aventail™ Link Server on supported provider servers, enrol it in Aventail™ Control and deploy the supported models your service will offer.
A managed service, not only capacity.
Give enterprise customers a clear path from model choice to useful work, on your compute and on the infrastructure they approve. They don’t need to be MLOps specialists to get started.
Reach their workflows and infrastructure
Customers call your endpoints directly or route from their own machines running Aventail™ Link, where their policy can keep some work on their infrastructure.
Serve air-gapped customer workloads
Deliver AI services inside a customer’s disconnected environment, using Aventail™ Link to run installed, supported models and locally deployed enterprise Aventail™ Control for management. Sensitive workflows can stay isolated while separate, connected workloads use your hosted compute where permitted.
Agree installation, updates, access and support with the customer. Network isolation and the complete application stack must be validated for each deployment; external endpoints are unavailable inside a true air gap.
See how the service is used
Review machines, models and routing events in Aventail™ Control, with Aventail™ support for deployment and operation.
Central management. Direct inference.
Aventail™ Control distributes models, policies and configuration to Aventail™ Link Server, then receives device and routing events. Inference requests reach the runtime directly. Providers continue to operate their infrastructure, with Aventail™ support agreed for the deployment.
Serve the customer’s estate, not only your racks.
Your Aventail™ Link Server can be one approved destination alongside the customer’s own devices, servers, private cloud and frontier APIs. The customer’s policy selects one destination for each request.
Customer deviceAVENTAIL™ LINK RUNTIME
On-premises machineAVENTAIL™ LINK RUNTIME
Private-cloud machineAVENTAIL™ LINK RUNTIME
Your inference endpointAVENTAIL™ LINK SERVER
Frontier APICUSTOMER’S PROVIDER ACCOUNT
Customers choose which destinations their policy permits: their own infrastructure, approved providers or your capacity. Keeping sensitive AI requests within approved infrastructure. How to reduce AI provider dependency.
Practical provider questions.
What does Aventail™ Server provide?
Aventail™ Server is a managed AI service offering for data centres and compute providers. It combines the Aventail™ Link Server runtime, model deployment, endpoints, application access, policy-based routing across your compute and the customer’s approved infrastructure, and central management, supported by the Aventail™ team. Aventail™ Link Server executes models; Aventail™ Control manages configuration and activity outside the inference request path.
Can Aventail™ Link Server replace vLLM?
Yes, for supported models and hardware. Before migrating, assess model formats, quantisation, the APIs your customers require, your hardware and performance on your real workloads. We’ll review these with you.
Do customers need deep MLOps expertise?
The experience is designed to help customers deploy, configure and route supported models without being MLOps specialists. Providers still operate their infrastructure, with Aventail™ support agreed for the deployment.
Who manages the customer side?
You operate your infrastructure. Management, access and support for machines running Aventail™ Link on the customer’s infrastructure are agreed for each deployment.
How does it fit our current serving setup?
We’ll review your current serving setup, supported models and customer workflows, then map how Aventail™ Link Server, deployment, endpoints and routing fit the service you want to offer.
Which hardware and models are supported?
Hardware and model scope is discussed with the Aventail™ team for the provider’s infrastructure and intended customer workflows.
How does a pilot begin?
Start with one enterprise customer workflow, your hardware and a representative model, then review the experience required from deployment through routing and support.
Build your managed AI service on Aventail™.
Bring your serving setup, model requirements and an enterprise customer workflow. We’ll map runtime deployment, customer access, routing and operational support across both sides.
Choose a time with our team.