AI Workflow Templates
ResearchadvancedHITL requiredupdated 2026-08-11

Agentic Research Brief

A plan-search-synthesize agent loop that produces a cited research brief with an explicit confidence note.

#agent#research#search#synthesis
View source →

Tools in the pipeline

  • llm
    planner
    Decompose into 3-6 sub-questions.
  • retriever
    search
    Retrieve sources per sub-question.
  • llm
    synthesizer
    Write the brief with citations and a confidence note.
  • human
    analyst
    Accept or send back conflicts.
Stack: LLM · search · citation store

What it does

Answers an open research question by planning sub-questions, searching for each, and synthesizing a brief with citations and a stated confidence level. Flags where sources disagree instead of papering over it.

Pipeline

  1. Plan — decompose the question into 3–6 sub-questions.
  2. Search — retrieve sources per sub-question (web, docs, or internal).
  3. Extract — pull the relevant claim + citation from each source.
  4. Synthesize — write the brief, citing every claim.
  5. Assess — end with a confidence note and list unresolved conflicts.

Guardrails

Every non-obvious claim must carry a citation. When sources conflict, present both and say so. Cap search depth and total tool calls to bound cost. State the confidence level explicitly so readers can calibrate.

Structural dry-run

No model call. Validates required fields only.

Prefills the example payload for MCP adoption. Edit it, then run.

Related