Skip to main content

How LexiCo’s AI technology works

LexiCo builds technology to route AI requests, make context more manageable and verify content later. Follow a request from the application to the model and see where the proxy, STONE and TSP contribute.

Illustration of a Nordic workspace with a glass-walled server room

Four engineering principles

  • Model routing

    Use one gateway to select from the models supported by the integration.

  • Compact context

    Extract the information the next step needs to carry on.

  • Verifiable content

    Bind content to cryptographic evidence that can be checked later.

  • Clear data flows

    Map what LexiCo processes and what data a model provider receives.

O(1) Proxy

The proxy sits between the application and models from different providers. O(1) describes bounded lookup and routing work inside the proxy relative to request size. Total response time depends on factors such as data volume, network conditions and the model generating the answer.

Illustration of a gateway with multiple network connections
  • One entry point

    The application uses the proxy as its interface to the models it can access.

  • Measure the right part of the request

    Measure proxy routing separately from transfer, queueing and model generation to see where time is spent.

  • Alternative models

    A configured fallback can be used when the primary model is unavailable and the integration rules allow it.

What happens in the proxy?

The proxy checks the request, selects a route and determines whether to retrieve a previous answer or forward the request. STONE can compress context before a new model call.

Three connected modules illustrate validation, model selection and reuse
  • Validate the request

    Check access and verify that the request matches the format expected by the integration.

  • Select a route

    Use the available models and the routing rules configured for the integration.

  • Reuse or forward

    When Deflect finds a suitable previous answer, it can serve the request without a new model call.

STONE compression

STONE creates a more compact representation of working context. It extracts facts, conclusions and open questions for later steps to build on. How much can be compressed depends on the material and the task.

Illustration of unstructured material becoming a compact, structured data stream
  • Keep what the work needs

    Carry relevant facts, decisions and unresolved points between steps.

  • Reduce repetition

    Shorter context can reduce input tokens. The cost impact depends on the model and its pricing.

  • Evaluate quality on your own tasks

    Compare answers and token use against the full-context baseline before choosing a compression level for the workflow.

Trust Standard Protocol (TSP)

TSP binds a document or event to cryptographic evidence. During verification, the content can be compared with the recorded fingerprint to determine whether it has changed.

Illustration of evidence modules with time and key symbols
  • Content integrity

    A cryptographic fingerprint makes it possible to verify that content matches the recorded version.

  • Timing evidence

    Timing information in the evidence provides a basis for assessing when the version was recorded.

  • Audit documentation

    Use the evidence alongside process and accountability records to document what was captured.

Architecture and integration

LexiCo’s web platform uses ASP.NET Core and SQL Server. When an application uses the AI proxy, the integration can manage model selection without the application connecting directly to every provider.

Illustration of modular server racks and network links overlooking a fjord
  • Integration point

    Identify the models, formats and capabilities the workflow needs before connecting to the proxy.

  • Shared sign-in

    LexiCo uses OpenID Connect for sign-in across connected services.

  • Define data processing boundaries

    Map which steps run in LexiCo’s environment and which are performed by the selected model provider.

Security and data handling

LexiCo’s own operations and processing by model providers are separate parts of the data flow. Who processes data, where that happens and how long it is retained must be established for the model, service and configuration the organisation uses.

Illustration of an operations space with a separate server room and workspace
  • Access control

    Determine who and what needs access, and which permissions the workflow requires.

  • Provider-side data

    Verify processing locations, retention periods and terms for requests sent to an external model.

  • Traceable documentation

    Use TSP evidence where content integrity needs to be documented. Assess privacy and regulatory obligations across the whole workflow.

From core technology to products

LexiCo’s products address different needs, from continuity in agent workflows and fewer model calls to desktop tools and training. The product catalogue also lists the other services.

Illustration of a shared environment for technology, collaboration and training
  • ContextContinuity

    Carries working context forward across long agent sessions, interruptions and model changes.

  • Deflect

    Reuses relevant previous AI answers when a new request is semantically similar.

  • Lexi Desktop Command

    Brings work with multiple AI models together in a desktop tool.

  • LexiCo Academy

    Offers organisations training in areas including AI, regulation, and health and safety.

Find the right solution for your workflow

Compare the products or contact LexiCo to discuss model selection, integration and data handling.