Consensus Scholar Agent: Research Automation with GPT-5 and Responses API

Consensus has introduced Scholar Agent, a multi-agent system built on GPT-5 and the Responses API that enables researchers to synthesize peer-reviewed literature in minutes. By automating the planning, searching, reading, and analysis of academic papers, the platform reduces the time spent on literature review from weeks to minutes.

Multi-Agent Architecture for Research Synthesis

Scholar Agent transitions Consensus from a vertical search engine to an agentic assistant by employing a coordinated workflow of specialized agents. This modular design ensures that reasoning remains precise and minimizes hallucinations by limiting the scope of each agent.

The system utilizes four primary agents:

  • Planning Agent: Deconstructs the user's query and determines the necessary sequence of actions.
  • Search Agent: Scans the Consensus paper index, user-private libraries, and the citation graph.
  • Reading Agent: Interprets academic papers either individually or in batches.
  • Analysis Agent: Synthesizes the final results, determines the visual structure, and composes the output.

This architecture implements what the team calls "context engineering," where the system assembles a "research context pack"—a structured bundle of metadata and findings—before the final generation occurs. If no relevant studies meet the quality threshold, the system is programmed to decline to answer rather than fabricate information.

Technical Implementation via Responses API and GPT-5

Consensus migrated from the Chat Completions API to the Responses API to better support multi-agent routing, which improved cost efficiency and reliability. The team selected GPT-5 based on early evaluations showing it outperformed GPT-4.1, Sonnet 4, and Gemini 2.5 Pro in two critical areas: tool-calling accuracy and planning stability.

This shift allowed the development team to focus on mapping agent behaviors to actual research workflows rather than optimizing prompts.

Market Adoption and Clinical Application

Consensus adopted a direct-to-consumer model targeting students, faculty, and clinicians rather than selling through academic institutions. This strategy led to the platform reaching over 8 million researchers worldwide and achieving an 8x increase in revenue over the past year.

Key milestones in adoption include:

  • User Base: Rapid growth among graduate students, PhD candidates, and faculty.
  • Medical Mode: A new feature specifically designed for practitioners seeking clinical evidence.
  • Institutional Partnerships: The company recently signed the Mayo Clinic’s medical library.

Future Scalability

The platform's modular architecture is designed to allow the integration of new agents as model capabilities expand. Future planned capabilities include agents capable of replicating experiments, running statistical analyses, and generating figures.

Sources