Skip to main content

LinkedInKGAgent

What it is​

A factory and thin agent wrapper that builds an IntentAgent specialized for querying LinkedIn data stored in a knowledge graph. It:

  • Loads SPARQL query tools from the “templatable SPARQL query” module.
  • Adds vector-search tools to resolve person/company URIs via semantic similarity.
  • Returns an IntentAgent instance (LinkedInKGAgent) configured with a LinkedIn-focused system prompt and intents.

Public API​

Constants​

  • NAME: "LinkedIn_KG"
  • DESCRIPTION: Human-readable agent description.
  • AVATAR_URL: LinkedIn logo URL.
  • SYSTEM_PROMPT: System instructions template. Tool list is injected at runtime.
  • SUGGESTIONS: Empty list (list[str]).

Functions​

  • create_agent(agent_shared_state: Optional[AgentSharedState] = None, agent_configuration: Optional[AgentConfiguration] = None) -> IntentAgent
    • Creates and returns a configured LinkedInKGAgent.
    • Wires SPARQL tools and two vector-search tools:
      • linkedin_search_person_uri (collection: linkedin_persons)
      • linkedin_search_company_uri (collection: linkedin_companies)

Classes​

  • class LinkedInKGAgent(IntentAgent)
    • No additional behavior; inherits everything from IntentAgent.

Configuration/Dependencies​

This module relies on an existing Naas ABI runtime with modules/services available:

  • Naas ABI core

    • naas_abi_core.services.agent.IntentAgent (IntentAgent, AgentConfiguration, AgentSharedState, Intent, IntentType)
    • naas_abi_core.modules.templatablesparqlquery.ABIModule to provide SPARQL tools via get_tools(...)
    • naas_abi_core.services.vector_store.VectorStoreService for similarity search

  • Marketplace / Application

    • naas_abi_marketplace.applications.linkedin.ABIModule.get_instance().engine must be initialized and contain:
      • module key: "naas_abi_core.modules.templatablesparqlquery"
      • service: engine.services.vector_store

  • LLM / embeddings

    • Chat model: naas_abi_marketplace.ai.chatgpt.models.gpt_4_1_mini.model
    • Embeddings: langchain_openai.OpenAIEmbeddings(model="text-embedding-3-large")
    • Requires numpy, pydantic, langchain_core.tools.StructuredTool

Usage​

from naas_abi_marketplace.applications.linkedin.agents.LinkedInKGAgent import create_agent
from naas_abi_core.services.agent.IntentAgent import AgentSharedState

agent = create_agent(agent_shared_state=AgentSharedState(thread_id="demo"))

# How you invoke/run the agent depends on IntentAgent's interface in naas_abi_core.
# This module only constructs and returns the configured agent instance.
print(agent.name)

Caveats​

  • The factory assumes the LinkedIn ABIModule engine is already initialized and contains the required module/service; otherwise tool wiring will fail (including an assert on the SPARQL module type).
  • Two intents reference tool names not defined in the tool list built here (e.g., linkedin_search_connections_by_person_name, linkedin_get_connection_information, linkedin_search_email_address_by_person_uri). Successful execution depends on those tools existing in the broader runtime/tool registry.
  • Vector search tools require populated vector store collections named:
    • linkedin_persons
    • linkedin_companies