Skip to content

Companion¤

The Companion view enables you to interact with your data, graphs, vocabularies, resources and queries in a chat-like way.

Configuration Info¤

A specific configuration is required to activate this feature. Our deployment templates (helm charts and docker-compose configurations) provide tested configuration profiles for different LLM providers. See the link in the info box for details.

Configuration

See spring.ai.* configuration properties in Configuration of Servlet Container to learn how to enable and configure this feature. Explore uses Spring AI to connect and configure LLM provider, as such the Spring AI Chat Model API Configuration Reference applies.

For optimal results, we recommend using Anthropic or OpenAI frontier-level models configured through the native Spring AI provider setup.

Usage¤

Companion Menu Button

You can open the Chat with Companion from any explore module via the Companion button in the top right of the application header, next to the user menu.

Companion Chat

In the companion sidebar, use:

  • to increase the size of the sidebar
  • to decrease the size of the sidebar
  • to start a new conversation
  • the dropdown menu to select or search for a conversation
  • to add documents as context to the chat (supported formats include pdf, txt and md)
  • to configure the tools that the LLM can use to answer.
  • to use your microphone to dictate your questions
  • to submit your question

To maximize the relevance and accuracy of the generated answer, your question is automatically enriched based on the context of the current view (e.g. the query text, the graph and the resource IRI).

Tool selection¤

Companion Tool Selection

The companion chat uses various tools to interact with Corporate Memory. These tools are used autonomously by the configured LLM model to gather the relevant context needed to generate the best answer. If your use case or question requires you to limit the available tools (e.g. if you only want to use queries from the query catalogue and not generate any), you can configure the tools the model can use via the drop-down widget.

  • Query Catalog related

    query_list
    Lists all existing SPARQL queries on this corporate memory instance.
    query_details
    Retrieve all information about a certain query in the query catalog.
    query_execute_saved
    Executes a saved SPARQL query on this corporate memory instance. The query is identified by its IRI.

  • Query related:

    query_execute_json
    Executes a SPARQL SELECT query on this corporate memory instance. The query is provided as a string.

  • Introspection related:

    graph_tree
    Retrieves a tree of all the direct and transitive owl:imports for a given graph.
    graph_class_usage
    Shows how classes are connected in a particular graph to each other by different properties.
    graph_list
    Lists all existing RDF graphs on this corporate memory instance.
    graph_properties_list
    Lists all used RDF properties used in this graph.
    graph_classes_list
    Lists all used owl:Class etc. used in this graph.
    vocabularies_property_list
    All Properties defined in this knowledge graph by installed vocabularies. While all properties are returned, they can easily be grouped by either their explicit association with an ontology in the isDefinedBy (rdfs:isDefinedBy) field, or by the sourceGraph field, which contains the names of the graphs from which the type statements were loaded.
    vocabularies_class_list
    All Classes defined in this knowledge graph by installed vocabularies. While all classes are returned, they can easily be grouped by either their explicit association with an ontology in the isDefinedBy (rdfs:isDefinedBy) field, or by the sourceGraph field, which contains the names of the graphs from which the type statements were loaded.

  • Resource related

    resource_search
    Performs a lucene syntax full text search of labels and descriptions for ‘searchTerm’ and retrieve list of best matches. The context graph will restrict the search to its includes, or search the full graph. Pattern allows to define a SPARQL WHERE clause snippet, which defaults to {GRAPH ?contextGraph {?resource ?p ?o}}. Search will always be on the ?resource, so adjust the pattern, if needed.
    resource_get
    Retrieves the Concise Bounded Description (CBD) of a resource from the knowledge graph and returns it as JSON-LD.

Interactive answers¤

Enriched answers

Your answer may include elements that allow rich interaction with specific Corporate Memory views and modules:

Queries : links will open a (generated) query in the query module for inspection or reuse.

Business Knowledge Editor : links open the respective resource(s) as starting node(s) in the Business Knowledge Editor module for visual exploration.

Resources : Resources Links links will open the resource details view in EXPLORE > Knowledge Graph module.

Comments