Multi-Agent Research Assistant

A production-quality multi-agent system built with LangGraph that automates research, writing, and editorial review through coordinated AI agents.

Architecture

flowchart LR
    SUP((Supervisor))

    subgraph Agents
        RES["Researcher\n(web search + summarize)"]
        WRI["Writer\n(draft / revise)"]
        REV["Reviewer\n(critique + verdict)"]
    end

    SUP -->|route| RES
    SUP -->|route| WRI
    SUP -->|route| REV
    RES -->|notes| SUP
    WRI -->|draft| SUP
    REV -->|verdict| SUP
    SUP -->|complete| FIN([FINISH])

Loading

Agent Roles

Agent	Role	Tools
Supervisor	Inspects shared state and routes work to the appropriate specialist agent. Decides when the task is complete.	None (LLM-based routing)
Researcher	Gathers information from the web relevant to the user's query and distils it into structured notes.	web_search (Tavily), summarize
Writer	Transforms research notes into a polished, well-structured report. Incorporates reviewer feedback on revisions.	LLM generation
Reviewer	Critiques the draft for accuracy, clarity, and completeness. Issues an ACCEPT or REVISE verdict.	LLM evaluation

Key Patterns

• Supervisor routing -- a central coordinator node uses conditional edges to dispatch work, avoiding brittle hard-coded pipelines.
• Human-in-the-loop -- the graph can be paused before the reviewer node using LangGraph's interrupt_before mechanism, allowing a human to inject feedback.
• Tool use -- the researcher agent calls external tools (web_search, summarize) to gather and condense information.
• Multi-provider LLM -- switch between OpenAI and Anthropic with a single environment variable.
• Iterative refinement -- the writer-reviewer loop runs up to 3 revision cycles, ensuring output quality.

Quick Start

# Clone the repository
git clone https://github.com/<your-username>/langgraph-multi-agent.git
cd langgraph-multi-agent

# Set up environment
cp .env-template .env
# Edit .env with your API keys

# Install dependencies with uv
uv sync

# Run a research query
uv run python main.py "What are the latest breakthroughs in quantum computing?"

# Verbose mode (full agent output)
uv run python main.py --verbose "Explain the current state of nuclear fusion energy"

Environment Variables

Variable	Required	Description
OPENAI_API_KEY	Yes (if LLM_PROVIDER=openai)	OpenAI API key
ANTHROPIC_API_KEY	Yes (if LLM_PROVIDER=anthropic)	Anthropic API key
LLM_PROVIDER	No	openai (default) or anthropic
TAVILY_API_KEY	Yes	Tavily API key for web search

Example Usage

$ uv run python main.py "Compare React and Svelte for building modern web apps"

============================================================
  Research query: Compare React and Svelte for building modern web apps
============================================================

--- [SUPERVISOR] ---
[Supervisor] Routing to: researcher

--- [RESEARCHER] ---
[Researcher] Gathered notes:
- React uses a virtual DOM; Svelte compiles to vanilla JS at build time
- React has a larger ecosystem and job market
- Svelte offers smaller bundle sizes and simpler syntax
...

--- [SUPERVISOR] ---
[Supervisor] Routing to: writer

--- [WRITER] ---
[Writer] Draft produced (2847 chars)

--- [SUPERVISOR] ---
[Supervisor] Routing to: reviewer

--- [REVIEWER] ---
[Reviewer] Verdict: ACCEPT
...

============================================================
  FINAL REPORT
============================================================

## React vs. Svelte: A Comparative Analysis
...

Docker

docker build -t research-assistant .
docker run --env-file .env research-assistant "Your research query here"

Tech Stack

Component	Technology
Orchestration	LangGraph
LLM (OpenAI)	GPT-4o via langchain-openai
LLM (Anthropic)	Claude Sonnet 4.5 via langchain-anthropic
Web Search	Tavily via langchain-community
Configuration	python-dotenv + pydantic
Build System	Hatch
Package Manager	uv
Containerisation	Docker (Python 3.12 slim)

Project Structure

langgraph-multi-agent/
├── agents/
│   ├── __init__.py
│   ├── config.py      # Multi-provider LLM configuration
│   ├── graph.py       # LangGraph StateGraph definition
│   └── tools.py       # Custom tools (web search, summarize)
├── main.py            # CLI entry point
├── pyproject.toml     # Project metadata and dependencies
├── Dockerfile         # Container build
├── .env-template      # Environment variable template
└── .gitignore

License

MIT