Data control
Keep inference inside the network.
Process selected inputs locally, close to the systems your application serves. Configure models, retrieval and dependencies locally when the project requires an offline workflow.
For ISVs, system integrators and developers
Run 30B-class AI.
On your own network.
Empato TrustedToken Ultra brings supported AI models to your application, on your network or your customer's. Combine 64GB unified memory, local AI compute and an OpenAI-compatible API to build local AI into the software you deliver.
30B-classSelected model configurations*
64GBLPDDR5x unified memory
400 TOPSlow-precision sparse compute
Local-firstOffline-capable configurations
Build for real deployment requirements
Your customers need AI.
Their data has boundaries.
Give customers a local deployment option when data policies, connectivity or recurring API usage shape the project. Put supported inference inside their environment and keep your application at the center.
Deployment economics
Put repeat AI work on local compute.
Run supported tasks locally instead of sending every request to a cloud model. Compare hardware, power and operating costs with API spend at the quality and usage your customers need.
Software integration
Add a local endpoint to your app.
Connect through an OpenAI-compatible API and supported Linux tooling. Keep your business logic in your application, with model and runtime compatibility checked for your stack.
Building local AI into a product or customer project?
Contact SalesApplications your team can build
Bring local AI into customer workflows.
Use Empato TrustedToken Ultra as the inference layer behind your software. Connect approved data, business rules and interfaces to build applications for the environments you serve.
Explore by industry
Choose an industry to explore an application concept. Swipe to see more.
Application conceptIllustrative workflow
Industrial automation
Put equipment knowledge inside your app.
Build a maintenance assistant that retrieves approved manuals and equipment records, then drafts source-linked answers for technicians to review.
- Inputs
- Equipment manuals, asset IDs and maintenance records
- Local AI task
- Retrieve relevant passages and draft a cited answer
- Customer benefit
- Find the procedure behind a fault question
Application conceptIllustrative workflow
Smart facilities
Connect service teams to site knowledge.
Add a local knowledge assistant to your facilities platform. Retrieve approved procedures and building information within the access rules you configure for each site.
- Inputs
- Site procedures, building records and service guides
- Local AI task
- Retrieve permitted content and draft a response
- Customer benefit
- Give service teams a shared source for recurring questions
Application conceptIllustrative workflow
Healthcare operations
Make operational procedures easier to find.
Build a local assistant for approved non-clinical SOPs and equipment guidance. Connect staff questions to the relevant source, with access controls defined by your application.
- Inputs
- Non-clinical SOPs and equipment service manuals
- Local AI task
- Retrieve approved guidance with source references
- Customer benefit
- Help staff locate operational information for review
Application conceptIllustrative workflow
Water management
Keep asset history in the maintenance workflow.
Connect asset records and approved manuals to your maintenance software. Generate a source-linked brief for an operator's question without making the model a process controller.
- Inputs
- Asset registers, maintenance logs and approved manuals
- Local AI task
- Retrieve asset-specific records and draft a brief
- Customer benefit
- Review the asset, its history and its procedure together
Application conceptIllustrative workflow
Energy operations
Bring the right context into the next shift.
Add a local handover assistant to your asset-management software. Use approved work orders and logs to draft a structured summary that operators can check against its sources.
- Inputs
- Work orders, maintenance logs and asset documents
- Local AI task
- Draft a source-linked summary of relevant records
- Customer benefit
- Carry open work and supporting context into handovers
These are application concepts, not preinstalled solutions or customer case studies. Connectors, access rules and model performance are configured for each project. Healthcare examples are non-clinical. Operational and safety decisions remain with qualified people and existing control systems.
Hardware and software, together
Local compute.
A stack your app can use.
Empato TrustedToken Ultra combines the device, a supported model runtime and API integration support. Your team builds the application and business logic around that foundation.
01 Compute
A local target for your model.
local AI compute and 64GB unified memory support selected 30B-class configurations. A compact enclosure places that inference capacity at the customer site.
Local AI compute + unified memory02 Runtime
A supported model stack.
Model selection, quantization and adaptation help prepare the runtime for deployment. Linux and supported model tooling provide a familiar foundation, with versions matched to the delivered configuration.
Model + quantization + runtime
03 Application
Connect through a local API.
Send requests from your application to the supported model endpoint. Keep access rules, retrieval, tool permissions and human approvals in the workflow your team controls.
Local endpoint + your business logicIncluded models, runtime versions, integration support and service scope are specified for the delivered configuration.
Build on the local inference layer
Your application.
Local AI inside.
Add document answers, voice interfaces, visual understanding or structured task outputs to your software. Each application combines a supported model with the data, tools and permissions your team provides.
Explore application types
Document AI
Turn approved documents into answers in your app.
Connect retrieval and a supported local language model to manuals, policies or technical records. Return source-linked answers through your interface, using the document permissions your application enforces.
Ask About IntegrationInputManuals + records
Local inferenceEmpato TrustedToken UltraOn the local network
OutputCited answer
Empato TrustedToken Ultra
Local AI compute.
In a compact device.
Place a supported inference stack on the customer network. Linux tooling, wired Ethernet and 512GB UFS 4.0 storage with M.2 NVMe SSD expansion provide the foundation for running models alongside your application.
View SpecificationsDevice viewsChoose a view
Front view · gray recessed top and muted bronze grille
Product renders show the enclosure. Rear ports are not pictured. Hardware revision and available connections are specified with the configuration.
The specifications behind the deployment
Built for local inference.
Sized for your application.
local AI compute architecture, a 12-core ARMv9.2 CPU and 64GB LPDDR5x unified memory. Model precision, context length and concurrent demand determine how that capacity serves your application.
Empato TrustedToken Ultra
Local compute with a supported model runtime and API.
64GB
LPDDR5x unified memoryShared by CPU and GPU for model weights, context, runtime and system use.
400TOPS
low-precision sparse computeLow-precision hardware throughput. Application response speed depends on the model and runtime.
30B-class
Selected model configurationsRun supported larger models locally with the required quantization and adaptation.
10+streams
Selected models up to 4BA separate small-model serving configuration for concurrent requests.
512GB UFS 4.0 storage with M.2 NVMe SSD expansion2.5GbE Ethernet35–150W configurable power
Ask About Model Support
* 30B-class capacity and 10+ streams describe different configurations. 400 TOPS is a low-precision sparse hardware rating, not tokens per second. Memory is shared with the operating system and runtime. Actual speed, quality, context and concurrency depend on the selected configuration. Power varies with configuration and load.
Built to connect with your software
Keep your application.
Add local inference.
Use an OpenAI-compatible API to connect your application to a supported local model. Linux and supported PyTorch tooling give your team a familiar starting point, with dependencies matched to the hardware and runtime.
- Connect your app to a local inference endpoint.
- Configure streaming, structured outputs and tools where supported.
- Keep data access and action permissions under your application's control.
From your app to a local modelExample API flow
Compatibility is model- and runtime-specific. Confirm the endpoints, schemas, tool behavior and package versions used by your application. OpenAI-compatible describes the API interface.
Built for the teams delivering AI
Your software expertise.
A local AI foundation.
You bring the application, customer knowledge and business logic. Empato TrustedToken Ultra provides the local compute and supported model stack to build on.
Independent software vendors
Add an on-premise deployment option.
Extend your product to customers who need inference inside their own environment. Keep the application experience yours while using a supported local model endpoint.
System integrators
Connect AI to the customer's systems.
Build around a defined compute platform and supported runtime. Focus integration work on the customer's data, interfaces and operational requirements.
Developers
Build with a local model API.
Use supported Linux and supported model tooling to develop against the selected model stack. Tune prompts, retrieval and tools around the application you are building.
Contact Sales
Models, software versions, service scope, pricing and availability are confirmed for the configuration you select.
Technical questions
Know what fits.
Know what ships.
The details that matter when you are building a product, integrating a customer site or choosing a local inference platform.
Model supportIntegrationDeployment
Ask a QuestionWhat comes with Empato TrustedToken Ultra?+
The platform combines local compute hardware with a supported model runtime and API integration options. The hardware specification includes local AI compute architecture, a 12-core ARMv9.2 CPU, 64GB LPDDR5x unified memory, 512GB UFS 4.0 storage with M.2 NVMe SSD expansion and 2.5GbE Ethernet. Confirm slot occupancy, delivered OS image, model configuration and included support in the quotation. Applications and additional devices are project-specific.
Can the application run completely offline?+
Yes, when the full application path is configured locally. Models, retrieval, authentication, tools and other required services must be available without external connections. Cloud APIs or remote services still require connectivity and may transmit data. A local inference endpoint alone does not make every connected application offline.
Which models and runtimes can I use?+
The supplied model range includes selected Qwen and Gemma configurations. Runtime options include vLLM, SGLang and llama.cpp where supported. Compatibility depends on the exact checkpoint, quantization, hardware support and runtime version. Selected 30B-class models and 10+ concurrent streams on selected models up to 4B are separate configurations. Confirm your model, context length and concurrency before choosing the setup.
Does local AI cost less than cloud inference?+
It can, particularly for suitable recurring workloads, but the comparison depends on usage and required quality. Include hardware, electricity, operations, support and any cloud calls or paid services. Locally processed requests do not use a third-party cloud inference API; that does not make local compute free. Empato also describes routing, gateway, data-retention and model-optimization services, with scope and fees confirmed separately.
Can I connect my existing application?+
An OpenAI-compatible API provides a familiar integration interface. Your team still checks the endpoints, request formats, streaming, structured outputs and tool calls the selected model and runtime support. Linux and supported model tooling do not guarantee every package or application runs unchanged. On the ARM platform, architecture-specific dependencies also need checking.
What are the power, warranty and purchase terms?+
The specified power range is 35–150W depending on configuration and load. The product brochure lists IP54 protection. The specified hardware warranty is 1 year. Confirm applicable regional terms, exclusions, pricing, included services and delivery availability with Empato. IP54 is not a waterproof rating.
Is the complete business application included?+
Empato TrustedToken Ultra is the local inference foundation. Your ISV or integration team provides the application, business rules, data connections and user experience. Document retrieval, voice, vision and agent workflows require the appropriate models and integrations. Additional devices, custom development and model optimization are included only when specified in the agreed scope.
Contact Sales
Build with
Empato TrustedToken Ultra.
Contact our team about model support, API integration, configuration and pricing for your product or customer deployment.
Empato TrustedToken Ultra inquiryStart with your company.
Three short steps
- Company
- Application
- Deployment
Please do not include passwords, customer records or confidential documents. Sending an inquiry does not place an order.