Skip to main content

ArXivAgent

What it is​

  • A thin wrapper around the core Agent that configures an ArXiv-focused assistant.
  • Bundles ArXiv search/metadata tools, a paper ingestion pipeline (knowledge graph + PDF download), and a query workflow for stored papers.

Public API​

  • create_agent(agent_shared_state: AgentSharedState | None = None, agent_configuration: AgentConfiguration | None = None) -> Agent

    • Creates and returns a configured ArXivAgent instance with:
      • Chat model: naas_abi_marketplace.ai.chatgpt.models.gpt_4_1_mini.model
      • Tools from:
        • ArXivIntegration.as_tools(...)
        • ArXivPaperPipeline(...).as_tools()
        • ArXivQueryWorkflow(...).as_tools()
      • Default AgentConfiguration(system_prompt=SYSTEM_PROMPT) if none provided
      • Default AgentSharedState() if none provided

  • class ArXivAgent(Agent)

    • Specialized agent class for ArXiv usage (no additional behavior beyond Agent).

Configuration/Dependencies​

  • Depends on naas_abi_core.services.agent.Agent:
    • Agent, AgentConfiguration, AgentSharedState
  • Initializes the ArXiv application module:
    • from naas_abi_marketplace.applications.arxiv import ABIModule
    • Uses module.engine.services.triple_store
  • Tooling components:
    • ArXivIntegration + ArXivIntegrationConfiguration
    • ArXivPaperPipeline + ArXivPaperPipelineConfiguration
      • storage_base_path="storage/triplestore/application-level/arxiv"
      • pdf_storage_path="datastore/application-level/arxiv"
    • ArXivQueryWorkflow + ArXivQueryWorkflowConfiguration
      • storage_path="storage/triplestore/application-level/arxiv"
  • Built-in prompt:
    • SYSTEM_PROMPT describes expected tool usage (search first, then fetch details, optionally ingest to KG, then query KG).

Usage​

from naas_abi_marketplace.applications.arxiv.agents.ArXivAgent import create_agent

agent = create_agent()

# Interact using the base Agent interface (method names depend on Agent implementation).
# Example (pseudo):
# response = agent.run("Find recent papers about retrieval-augmented generation and summarize trends.")
# print(response)

Caveats​

  • Requires the ArXiv ABIModule engine and its triple_store service to be available at runtime.
  • Tool availability and how to invoke the agent depend on the underlying Agent implementation (this file only wires configuration and tools).