Research & Analysis Agent Blueprint

What This Blueprint Does

The Research Agent is a three-stage pipeline: gathering information from designated sources, synthesizing findings into coherent analysis, and formatting structured reports. Every source is attributed, every claim is traceable, and the process is logged through Sentrely.

This is not a scraper that dumps raw text. The agent evaluates source credibility, cross-references claims, identifies contradictions, and flags areas where evidence is thin.

Architecture

Three specialized agents with distinct permissions:

1. Research Agent — Gathers from approved web domains and search. Has scoped web access only.
2. Synthesis Agent — Cross-references findings, identifies patterns. No web access — works only on gathered data.
3. Formatter Agent — Produces the final report. No web or data access beyond synthesis output.

Only the Research Agent can access the web, and only within approved domains. The other agents are fully sandboxed.

Policy Configuration

project: acme-research
agent: research-pipeline

policies:
  # Scoped web access — add domains you trust
  - web:read on domains/arxiv.org
  - web:read on domains/scholar.google.com
  - web:read on domains/*.gov
  - web:search on engines/google
    max_queries_per_session: 50

  # Store research outputs
  - aws:s3:PutObject on arn:aws:s3:::acme-research/outputs/*
  - aws:s3:GetObject on arn:aws:s3:::acme-research/outputs/*

budget:
  max_tokens_per_session: 200000
  max_sessions_per_day: 10

notifications:
  on_complete: slack:#research-reports

Source Attribution

Every claim in the final report includes inline attribution with URLs and access timestamps. The bibliography makes it easy to verify claims and trace sources back to their origin.

Customizing for Your Domain

Use Case	Domains	Sources
Market research	Financial data providers, industry associations	Analyst reports, SEC filings
Technical research	arxiv.org, ACM, GitHub	Papers, RFCs, documentation
Competitive intelligence	Company domains, patent DBs, news	Websites, job postings, press releases
Regulatory	.gov domains, legal databases	Government publications, filings

Why Scoped Web Access Matters

Unscoped AI agents with web access can access internal URLs or gather data from unreliable sources. The Gateway domain allowlist ensures the agent only visits approved domains, all queries are logged and auditable, and source quality is controlled by policy, not agent judgment.