Advanced Examples

Preface

As advanced examples, we are going to use the following tutorials from LangGraph:

• Agent Architectures - Multi-Agent Network

• Agent Architectures - Multi-Agent Supervisor

• Agent Architectures - Hierarchical Agent Teams

We are also not going to focus on a basic usage - the main focus is going to be on langgraph_compare.jsons_to_csv.GraphConfig.

If You want to check basic usage, refer to: Getting Started.

Multi-Agent Network

This example is based on Agent Architectures - Multi-Agent Network and expands on Exporting JSON’s to CSV from Getting Started.

Multi-Agent Network diagram - LangGraph Documentation

The premise of this example is to show that GraphConfig can have multiple nodes.

In case of this example - we will have 2 nodes - researcher, chart_generator.

Example:

# Needed imports
from langgraph_compare.experiment import create_experiment
from langgraph_compare.jsons_to_csv import GraphConfig, export_jsons_to_csv

# Init for experiment project structure
exp = create_experiment("test")

# We can add multiple nodes!
graph_config = GraphConfig(
    nodes=['researcher','chart_generator']
)

# You can provide You own file name as an optional attribute csv_path.
# Otherwise it will use the default file name - "csv_output.csv"
export_jsons_to_csv(exp, graph_config)

Multi-Agent Supervisor

This example is based on Agent Architectures - Multi-Agent Supervisor. It introduces the concept of a Supervisor - a node that controls other nodes.

Multi-Agent Supervisor diagram - LangGraph Documentation

In this example, we will introduce langgraph_compare.jsons_to_csv.SupervisorConfig. It will supervise the graph - working more or less work the same as GraphConfig. The concept of supervisors will make more sense in Hierarchical Agent Teams.

Example:

# Needed imports
from langgraph_compare.experiment import create_experiment
from langgraph_compare.jsons_to_csv import GraphConfig, SupervisorConfig, export_jsons_to_csv

# Init for experiment project structure
exp = create_experiment("test")

# Supervisor for graph
supervisor = SupervisorConfig(
    name="supervisor",
    supervisor_type="graph"
)

# Config with supervisor and additional nodes
graph_config = GraphConfig(
    supervisors=[supervisor],
    nodes=["Researcher", "Coder"]
)

# You can provide You own file name as an optional attribute csv_path.
# Otherwise it will use the default file name - "csv_output.csv"
export_jsons_to_csv(exp, graph_config)

Hierarchical Agent Teams

This example is based on Agent Architectures - Hierarchical Agent Teams. It introduces the concept of a SubgraphConfig - a node that controls other nodes.

Hierarchical Agent Teams diagram - LangGraph Documentation

In this example, we have a Graph that is build from two SubGraphs. Those graphs are controlled by a Supervisor - that routes traffic to subgraphs. Furthermore, every graph has its own supervisor - that controls what is happening inside of it.

IMPORTANT: Be sure to call supervisors with different names - so you can differentiate between them! Calling supervisors with the same names WILL brake the parser.

Example:

# Needed imports
from langgraph_compare.experiment import create_experiment
from langgraph_compare.jsons_to_csv import GraphConfig, SubgraphConfig, SupervisorConfig, export_jsons_to_csv

# Init for experiment project structure
exp = create_experiment("test")

# Config for entire graph supervisor
graph_supervisor = SupervisorConfig(
    name="graph_supervisor",
    supervisor_type="graph"
)

# Config for Research Team subgraph supervisor
research_supervisor = SupervisorConfig(
    name="research_supervisor",
    supervisor_type="subgraph"
)

# Config for Paper Writing Team subgraph supervisor
paper_supervisor = SupervisorConfig(
    name="paper_supervisor",
    supervisor_type="subgraph"
)

# Config for Research Team subgraph
research_team = SubgraphConfig(
    name="ResearchTeam",
    nodes=["Search", "WebScraper"],
    supervisor=research_supervisor
)

# Config for Paper Writing Team subgraph
paper_team = SubgraphConfig(
    name="PaperWritingTeam",
    nodes=["DocWriter", "NoteTaker","ChartGenerator"],
    supervisor=paper_supervisor
)

# Config for complete graph
graph_config = GraphConfig(
    supervisors=[graph_supervisor],
    subgraphs=[research_team, paper_supervisor]
)

# You can provide You own file name as an optional attribute csv_path.
# Otherwise it will use the default file name - "csv_output.csv"
export_jsons_to_csv(exp, graph_config)

Notice how:

• every supervisor has a config (both graph and subgraphs) - but they have a different supervisor_type.

• every graph has a config (both graph and subgraphs) - but they are using different classes: GraphConfig or SubgraphConfig

• GraphConfig doesn’t have nodes defined - since they are being taken care of by subgraphs.