AgentRecommendationWorkflow

What it is

A Workflow implementation that:

• Loads templated SPARQL queries from a Turtle (.ttl) file.
• Matches a natural-language intent to a query template.
• Templates and executes the query against an Oxigraph SPARQL endpoint.
• Formats results into a list of agent recommendations.
• Exposes the workflow as a LangChain StructuredTool.

Public API

Classes

• AgentRecommendationConfiguration(WorkflowConfiguration)

  • Holds runtime dependencies and file/endpoint locations.
  • Fields:
    • triple_store: ITripleStoreService (present but not used in this module)
    • oxigraph_url: str (default: http://localhost:7878)
    • queries_file_path: str (default: src/core/modules/abi/ontologies/application-level/AgentRecommendationSparqlQueries.ttl)

• AgentRecommendationParameters(WorkflowParameters)

  • Input parameters for recommendations.
  • Fields:
    • intent_description: str (required)
    • min_intelligence_score: Optional[int] = 10
    • max_input_cost: Optional[float] = None
    • max_results: Optional[int] = 10
    • provider_preference: Optional[str] = None

• AgentRecommendationWorkflow(Workflow)

  • Main workflow class.
  • __init__(configuration: AgentRecommendationConfiguration)
    • Loads SPARQL query templates from queries_file_path.
  • run_workflow(parameters: WorkflowParameters) -> Dict[str, Any]
    • Validates parameter type (AgentRecommendationParameters only).
    • Selects a query by keyword matching on intent_description.
    • Replaces {{ var }} placeholders and processes {% if var %}...{% endif %} blocks.
    • Executes query via HTTP POST to "{oxigraph_url}/query".
    • Returns:
      • intent, query_used, recommendations, total_found
  • as_tools() -> list[BaseTool]
    • Returns a LangChain StructuredTool named recommend_ai_agents that calls run_workflow.
  • as_api(...) -> None
    • Present but currently does nothing and always returns None.
  • get_configuration() -> AgentRecommendationConfiguration
    • Returns the workflow configuration.

Configuration/Dependencies

Runtime dependencies

• SPARQL endpoint: Oxigraph accessible at configuration.oxigraph_url (expects POST /query returning JSON in SPARQL results format).
• Query templates file: configuration.queries_file_path must exist and be parseable as Turtle.
  • Extracts entries where the graph has subjects with predicate INTENT_MAPPING.intentDescription.
  • For each such subject, it reads:
    • INTENT_MAPPING.intentDescription
    • INTENT_MAPPING.sparqlTemplate

Python packages used

• requests (HTTP calls)
• rdflib (parsing TTL query file)
• langchain_core.tools (tool wrapper)

Usage

Run the workflow directly

from naas_abi.workflows.AgentRecommendationWorkflow import (
    AgentRecommendationWorkflow,
    AgentRecommendationConfiguration,
    AgentRecommendationParameters,
)
 
# triple_store is required by the configuration type but not used by this workflow.
dummy_triple_store = object()
 
wf = AgentRecommendationWorkflow(
    AgentRecommendationConfiguration(
        triple_store=dummy_triple_store,  # must be provided
        oxigraph_url="http://localhost:7878",
        queries_file_path="path/to/AgentRecommendationSparqlQueries.ttl",
    )
)
 
out = wf.run_workflow(
    AgentRecommendationParameters(
        intent_description="Need help writing a business proposal",
        min_intelligence_score=50,
        max_results=5,
    )
)
 
print(out["total_found"])
print(out["recommendations"][:1])

Use as a LangChain tool

tools = wf.as_tools()
result = tools[0].invoke({"intent_description": "fast coding assistant", "max_results": 3})
print(result["recommendations"])

Caveats

• The .ttl queries file must exist; otherwise initialization raises FileNotFoundError.
• Intent matching is keyword-based with a fixed mapping; unmatched intents default to the “business proposal” query.
• Template substitution is plain string replacement for {{ key }} plus a basic regex-based {% if var %} handler (no loops/advanced templating).
• as_api(...) is a stub and does not register any routes.
• triple_store is required in configuration but is not used in this module.