Research report

gen_research_report · slug research-report · icon 🔍 · outputs markdown, docx

A deep, cited report on any question. The researcher breaks the question into sub-questions, runs a web research pass on each, and an analyst synthesises the findings, notes where sources disagree, and appends every source. Markdown and Word.

Inputs

Field	Type	Notes
Research question	text	required, 10-600 chars — one clear question beats three vague ones
Scope and context	text	optional — geography, time frame, inclusions/exclusions
Depth	pick	Quick (4 sub-questions) · Standard (6) · Deep (8)
Written for	pick	A decision-maker · A technical reader · A general reader
Length	pick	Brief (~1,200 words) · Standard (~2,500) · Long (~4,500)
Prefer sources from	pick	Any time · The last year · The last month

Pipeline

Step	Role	Skills	Tier	Depends on
plan	researcher	llm.generate	cheap	—
research	researcher	web.research, web.search, web.fetch	cheap	plan
synthesis	analyst	llm.generate	standard	plan, research
publish	publisher	docs.write_markdown, docs.write_docx, files.write	cheap	synthesis

This is the only seed generator where “research” runs as its own dedicated model-decomposed step (plan breaks the question into sub-questions) rather than the planner’s automatic research-task insertion — the whole point of this generator is the decomposition itself, applying the same web.research skill at higher depth per sub-question.

Example run

{
  "generatorId": "gen_research_report",
  "inputs": {
    "question": "What are the realistic options for a small studio to run open-weight language models on its own hardware in 2026, and what do they cost?",
    "depth": "Standard (6 sub-questions)",
    "audience": "A decision-maker"
  }
}