🔬 Deep Research Agent

A CLI-based deep research AI agent built with LangChain, LangGraph, and free Groq AI.

It breaks down complex research queries into sub-questions, iteratively researches each one, analyzes gaps, and synthesizes a comprehensive report — all powered by Groq's blazing-fast inference on Llama models.

Architecture

START → Decompose → Research → Analyze → [loop?] → Synthesize → END
            │            │          │                     │
       Break query   Research    Check for          Write final
       into sub-     each sub-   gaps & decide      comprehensive
       questions     question    to loop or finish   report

Built as a LangGraph state graph with 4 nodes and a conditional loop edge.

Setup

# Install globally
npm install -g deep-research-agent

# Configure your free Groq API key
deep-research setup

Get your free API key at console.groq.com/keys (no credit card required).

Usage

# Basic research
deep-research "What are the latest trends in AI agents?"

# Deep research (more iterations)
deep-research "Compare React vs Vue vs Svelte in 2025" --depth deep

# Save report to file
deep-research "State of WebAssembly" --output report.md

# Use a specific model
deep-research "Quantum computing progress" --model llama-3.3-70b-versatile

Options

| Flag | Description | Default | |------|-------------|---------| | -d, --depth | Research depth: shallow, medium, deep | medium | | -o, --output | Save report to a file | — | | -m, --model | Groq model to use | llama-3.3-70b-versatile |

How It Works

1. Decompose — Breaks your query into 3-5 focused sub-questions
2. Research — Investigates each sub-question in depth using the LLM
3. Analyze — Reviews findings for gaps and contradictions
4. Loop — If gaps are found and depth allows, generates new sub-questions and researches again
5. Synthesize — Produces a structured report with executive summary, findings, insights, and conclusions

Tech Stack

• LangChain — LLM orchestration and prompt management
• LangGraph — Stateful graph-based agent workflow with conditional edges
• Groq — Free, ultra-fast LLM inference (Llama 3.3 70B)
• Commander — CLI framework
• Chalk + Ora — Terminal UI

License

MIT