Skip to main content

PubMedAgent

What it is​

  • A thin agent factory for a PubMed-focused Agent configured with:
    • A PubMed search pipeline exposed as tools.
    • A helper tool to download PubMed Central PDFs by PMCID and store them in object storage.
  • Includes constants for the agent name, description, and system prompt.

Public API​

  • Class PubMedAgent(Agent)

    • Subclasses naas_abi_core.services.agent.Agent.Agent.
    • No additional methods/attributes (defined as pass).

  • Tool function download_pdf(pmcids: List[str]) -> str

    • LangChain tool (@tool) to download one or more PDFs from PubMed Central using PMCIDs.
    • Behavior:
      • For each PMCID:
        • Downloads PDF bytes via PubMedIntegration(PubMedAPIConfiguration()).download_pubmed_central_pdf(pmcid).
        • Writes to a temporary file, then uploads to object storage at:
          • bucket/prefix: "pubmed/pdfs"
          • key: "{pmcid}.pdf"
        • Deletes the temporary file.
      • Runs downloads concurrently via ThreadPoolExecutor(max_workers=10).
    • Returns: "PDFs downloaded and saved."

  • Factory function create_agent() -> PubMedAgent

    • Builds a PubMedPipeline and converts it to tools via pipeline.as_tools().
    • Returns a PubMedAgent configured with:
      • name, description, system_prompt constants.
      • Chat model: naas_abi_marketplace.ai.chatgpt.models.gpt_4_1.model.model (cast to ChatModel).
      • Tools: pipeline tools + download_pdf.
      • State: AgentSharedState(thread_id=<random uuid hex>).
      • memory=None, agents=[].

Configuration/Dependencies​

  • External services

    • Object storage is accessed through ABIModule.get_instance().engine.services.object_storage and must be available/configured in the runtime.

  • PubMed integration

    • Uses PubMedIntegration + PubMedAPIConfiguration to download PDFs from PubMed Central.

  • Pipeline

    • PubMedPipeline(PubMedPipelineConfiguration()) is used to provide additional tools (not defined in this file).

  • Model

    • Uses the gpt_4_1 chat model wrapper imported as model.

Usage​

from naas_abi_marketplace.applications.pubmed.agents.PubMedAgent import create_agent, download_pdf

agent = create_agent()

# Use the tool directly (requires ABIModule/object_storage to be configured in your environment)
result = download_pdf(["PMC1234567", "PMC7654321"])
print(result)

Caveats​

  • download_pdf prints "Downloading {pmcid}" to stdout for each PMCID.
  • Files are downloaded concurrently (up to 10 workers); failures/exceptions during download or storage upload are not handled in this function.
  • The agent’s system prompt requires displaying tool results as a Markdown table, but this file does not enforce formatting beyond the prompt text.