About the options
The controls above allow you to fine-tune how the Research Report Agent discovers, evaluates,
and synthesizes information. Most users can keep the default settings and still obtain
high-quality results.
- Mode – Determines the style of the report. Survey produces a broad, balanced overview of the topic, while other modes focus on more targeted positioning or framing.
- Recency policy – Controls how much emphasis is placed on recent versus foundational research. Balanced is recommended for most use cases.
- Literature search model – The language model used to retrieve and rank relevant research papers.
- Literature judge model (optional) – An additional model used to evaluate and refine literature relevance. Leaving this blank disables the judging step.
- Writer model – The model responsible for generating the final written report.
- Judge model – A separate model that reviews and scores report drafts to improve clarity, coherence, and quality.
- Max tool calls – Limits how many external research queries the agent can make during literature discovery.
- Top N total – The maximum number of research papers considered when constructing the report.
- Max iters – The number of internal revision cycles used to refine the final report.
Increasing limits or iterations may improve depth and quality but can increase processing time.
Defaults are chosen to balance insight, performance, and responsiveness.