Set your data boundaries
Choose where models can run and which destinations each workflow may use, including its fallback paths.
Open-weight models.
Frontier APIs.
Your hardware.
One orchestration layer.
Aventail™ is an integrated AI inference platform combining a built-in inference engine, model deployment across hardware, and policy-based inference routing.
Run supported models on your devices, servers and private cloud, with access to approved external AI providers. Deploy, run and manage inference through one platform, without separately assembling and maintaining a model-serving stack.
Install Aventail™.
Get the desktop app, or install the runtime from your terminal.
Intelligent AI orchestration.
Decisions made locally.
Secure on-device content-aware routing. Aventail™ Link evaluates the query, model fit, device conditions and your organisation’s requirements, then applies your rules to choose where inference runs.
Interactive illustration. Example signals and infrastructure; no query is sent.
Query
Help me with this request.
Query signalsContent context
Model fitCapabilities and task fit
Device stateSpecifications and availability
Organisation requirementsGeography, approvals and latency preference
Local policy checkRules evaluated on this device
Execution destinationAwaiting decision
- Ready
- Evaluate local context
- Check local policy
- Select destination
Ready
Set the example signals, then run the query.
Aventail™ Link will evaluate the query and constraints locally. No query leaves this illustration.
Explore the example environment
Example properties only. Default order: this device, another machine, on-premises server, private-cloud machine, provider server, frontier API. Lower-latency preference order: this device, another machine, frontier API, provider server, private-cloud machine, on-premises server. Approval is configured by the organisation, not a certification; latency ordering is illustrative, not measured. Hard requirements always win.
-
This deviceAventail™ Link
UK · Organisation-managed · Standard model · Not on example ESG-approved list
-
Another machineAventail™ Link
UK · Organisation-managed · Standard model · On example ESG-approved list
-
On-premises serverAventail™ Link
UK · Organisation-managed · Standard + advanced model · Not on example ESG-approved list
-
Private-cloud machineAventail™ Link
Outside UK · Organisation-managed · Standard + advanced model · On example ESG-approved list
-
Provider serverAventail™ Link Server
UK · External provider · Standard + advanced model · On example ESG-approved list
-
Frontier APICustomer’s provider account
Outside UK · External provider · Standard + advanced model · Not on example ESG-approved list
Content analysis and policy checking happen inside Aventail™ Link. Aventail™ doesn’t need the query to make the routing decision.
Aventail™ Control distributes configuration ahead of time. Prompts and responses do not pass through the hosted Aventail™ Control service. Approved destinations receive the requests routed to them; frontier API calls use your own provider account. See how Aventail™ Link Server works for inference providers.
Deploy the models.
Connect the machines.
Aventail™ brings model deployment, execution, routing and management into one system. Aventail™ Link runs on your machines. Aventail™ Control manages them. Workspace gives your teams a place to use AI.
Set the rules. See where work runs.
Manage machines, models and routing policies centrally. Distribute configuration to Aventail™ Link and review device and routing activity, outside the inference request path.
-
Deploy a model
Install Aventail™ Link on supported hardware and deploy the model you want to run.
-
Set the policy
Choose allowed models, destinations and fallback paths in Aventail™ Control.
-
Route and run
Aventail™ Link applies policy locally, then runs the request or sends it to an allowed destination.
-
Inspect the activity
Review the model, destination and routing rule in Aventail™ Control.
Control the data. Choose the compute.
Keep sensitive work on approved infrastructure. Use your own hardware where it fits, and reach frontier APIs when your policy allows.
Decide when to use paid APIs
Run suitable workloads on hardware you already operate. Choose when a frontier model is worth using and keep external calls on your own provider account.
See why a request went there
Review the model, destination and policy rule behind routing decisions, alongside device and routing events.
A practical start
Begin with one workflow on your own infrastructure. Expand to other devices, private-cloud machines and approved providers as your requirements grow. Read the guide to keeping sensitive requests within approved infrastructure.
Built by engineers.
Grounded in deployment.
Saif Al-Ibadi and Joseph Ward have built production AI together since 2018, from neuromorphic computing to generative design for aerospace.
That experience shaped Aventail™: AI has to work with the hardware, connectivity and constraints in front of it.
Featured in
Practical questions.
Direct answers.
How does Aventail™ differ from an AI gateway or LLM router?
An AI gateway or LLM router helps applications choose between model endpoints. Aventail™ also provides the runtime that deploys and executes supported models on your hardware. Aventail™ Link evaluates content and policy locally, then runs the request or routes it to another approved machine or frontier API. Aventail™ Control manages those machines outside the inference request path.
Can Aventail™ run on our own infrastructure?
Yes. Aventail™ Link runs on your devices and servers, and Aventail™ Control may be hosted by Aventail™ or in your own estate.
Does Aventail™ Control receive our prompts and responses?
The hosted Aventail™ Control service receives device and routing events, not prompts or responses. Aventail™ Link evaluates the applicable local policy on your machine.
Can we use frontier APIs and local models?
Yes. Policy can route to a model on this device, another machine running Aventail™ Link or a frontier API through your own provider account. Local execution can continue offline when the required model is installed and your policy allows it. Remote machines and APIs require a connection.
How do we get started?
Start with one useful workflow. Book a conversation with our team to discuss its requirements, your existing systems and the constraints that matter.
Put your infrastructure
to work for AI.
Bring one workflow, the hardware you have and the models you want to use. We’ll map where requests should run and what your first deployment needs.
Choose a time with our team.