# Component Interaction Map (/docs/coreai/component-interaction-map)



The CoreAI services are designed to work as a connected layer rather than as isolated deployments.

Interaction View [#interaction-view]

<Mermaid
  chart="flowchart LR
    PORTAL[&#x22;CoreAI Web Portal&#x22;] --> API[&#x22;CoreAI API&#x22;]
    PORTAL --> MI[&#x22;Model Installer&#x22;]
    PORTAL --> MLF[&#x22;MLflow&#x22;]
    PORTAL --> LLM[&#x22;LiteLLM&#x22;]

    API --> KC[&#x22;Keycloak&#x22;]
    API --> LLM
    API --> MIL[&#x22;Milvus&#x22;]
    API --> TEMP[&#x22;Temporal&#x22;]
    API --> MLF
    API --> OTEL[&#x22;OpenTelemetry&#x22;]
    API --> TH[&#x22;ToolHive&#x22;]

    TEMP --> DOC[&#x22;Docling&#x22;]
    TEMP --> MIL

    MI --> MLF
    MI --> LLM
    MI --> INF[&#x22;Runtime Inference&#x22;]

    DOC --> MIL
    LLM --> INF
    OTEL --> MLF"
/>

Key Patterns [#key-patterns]

* The CoreAI Web Portal is the main user surface and consumes several backend services.
* The CoreAI API acts as the central orchestration layer for application logic, retrieval, and AI integrations.
* Temporal sits behind parts of the CoreAI API workflow layer, especially asynchronous chat and file-processing paths.
* Model Installer bridges the model registry and the Runtime inference layer.
* LiteLLM acts as the shared LLM gateway for multiple CoreAI services.
* OpenTelemetry and MLflow provide the visibility layer for AI behavior and service telemetry.
* Docling is primarily involved through Temporal-driven document-processing workflows, while Milvus is used both for ingestion outputs and later retrieval.