Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/IconDean/research-agent/llms.txt

Use this file to discover all available pages before exploring further.

Deep Research Agent is an autonomous AI system that takes a natural-language research question and returns a structured, cited markdown report — without manual searching, reading, or summarizing. Powered by Google Gemini and DuckDuckGo, it runs a multi-step pipeline: planning sub-queries, fetching credible sources, detecting knowledge gaps, and synthesizing findings into a professional report with numbered citations.

Quickstart

Set up the backend and run your first research query in under five minutes.

Configuration

Configure API keys, model selection, and agent constraints.

Research Pipeline

Understand the five-stage agentic loop from planning to report synthesis.

REST API

Integrate the research agent into your own applications via HTTP.

What it does

The agent follows a five-stage research pipeline for every question:
1

Plan

Gemini decomposes the question into prioritized sub-queries and selects a search strategy (breadth-first or deep-dive).
2

First-round research

The agent searches DuckDuckGo and fetches the top pages, scoring each source for credibility before reading it.
3

Gap detection

Gemini reviews initial findings and identifies unanswered aspects, generating targeted follow-up queries.
4

Second-round research

The agent searches and fetches again — focused on filling the identified gaps.
5

Synthesis

All gathered content is compiled into a structured markdown report with an Executive Summary, Key Findings, Confidence Assessment, and a numbered source list.

Two ways to use it

CLI

Run python main.py "your question" from any terminal. Progress prints to stderr; the report prints to stdout.

Web UI

Start the FastAPI backend and the React frontend to get a live activity log and rendered report in your browser.

Key capabilities

Credibility Scoring

Sources are scored 0–1 based on domain authority, recency signals, and relevance. Low-scoring sources are blocked automatically.

Report Format

Every report follows a consistent structure: Methodology, Executive Summary, Main Findings, Conflicting Information, and Sources.

Python Modules

Use AgentRunner directly in your Python scripts for programmatic access to the full research pipeline.

Error Handling

Failed searches and fetches are logged and skipped — the agent continues and still produces a report.

Build docs developers (and LLMs) love