Using FalkorGraphRagCapability with agents for GraphRAG Question & Answering

Using FalkorGraphRagCapability with agents for GraphRAG Question & Answering

AG2 provides GraphRAG integration using agent capabilities. This is an example to integrate FalkorDB (a Knowledge Graph database).

FalkorDB’s GraphRAG-SDK is a dependency for this notebook, which can be installed with ag2 via pip:

pip install ag2[graph-rag-falkor-db]

or if you have already installed ag2/autogen/pyautogen

pip install graphrag_sdk==0.3.3

For more information, please refer to the installation guide.

Set Configuration and OpenAI API Key

By default, in order to use FalkorDB you need to have an OpenAI key in your environment variable OPENAI_API_KEY.

You can utilise an OAI_CONFIG_LIST file and extract the OpenAI API key and put it in the environment, as will be shown in the following cell.

Alternatively, you can load the environment variable yourself.

Learn more about configuring LLMs for agents here.

import os

import autogen

config_list = autogen.config_list_from_json(env_or_file="OAI_CONFIG_LIST")

# Put the OpenAI API key into the environment
os.environ["OPENAI_API_KEY"] = config_list[0]["api_key"]

Create a Knowledge Graph with Your Own Data

Note: You need to have a FalkorDB graph database running. If you are running one in a Docker container, please ensure your Docker network is setup to allow access to it.

In this example, the FalkorDB endpoint is set to host=“172.18.0.3” and port=6379, please adjust accordingly. For how to set up FalkorDB, please refer to https://docs.falkordb.com/

Below, we have some sample data from IMDB on the movie ‘The Matrix’. See the contents of the file here.

We then initialise the database with that text document, creating the graph in FalkorDB.

A Simple Example

In this example, the graph ontology is auto-generated. This allows you to load data without specifying the specific types of entities and relationships that will make up the database (however, this may not be optimal and not cost efficient).

from autogen import ConversableAgent, UserProxyAgent
from autogen.agentchat.contrib.graph_rag.document import Document, DocumentType
from autogen.agentchat.contrib.graph_rag.falkor_graph_query_engine import FalkorGraphQueryEngine
from autogen.agentchat.contrib.graph_rag.falkor_graph_rag_capability import FalkorGraphRagCapability

# Auto generate graph schema from unstructured data
input_path = "../test/agentchat/contrib/graph_rag/the_matrix.txt"
input_documents = [Document(doctype=DocumentType.TEXT, path_or_url=input_path)]

# Create FalkorGraphQueryEngine
query_engine = FalkorGraphQueryEngine(
    name="The_Matrix_Auto",
    host="172.17.0.4",  # Change
    port=6379,  # if needed
)

# Ingest data and initialize the database
query_engine.init_db(input_doc=input_documents)

# Create a ConversableAgent (no LLM configuration)
graph_rag_agent = ConversableAgent(
    name="matrix_agent",
    human_input_mode="NEVER",
)

# Associate the capability with the agent
graph_rag_capability = FalkorGraphRagCapability(query_engine)
graph_rag_capability.add_to_agent(graph_rag_agent)

# Create a user proxy agent to converse with our RAG agent
user_proxy = UserProxyAgent(
    name="user_proxy",
    human_input_mode="ALWAYS",
)

user_proxy.initiate_chat(graph_rag_agent, message="Name a few actors who've played in 'The Matrix'")

user_proxy (to matrix_agent):

Name a few actors who've played in 'The Matrix'

--------------------------------------------------------------------------------
matrix_agent (to user_proxy):

Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving are a few actors who've played in 'The Matrix'.

--------------------------------------------------------------------------------
user_proxy (to matrix_agent):

Any additional actors?

--------------------------------------------------------------------------------
matrix_agent (to user_proxy):

I'm sorry, but I don't have information about any additional actors from 'The Matrix' beyond Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving.

--------------------------------------------------------------------------------

ChatResult(chat_id=None, chat_history=[{'content': "Name a few actors who've played in 'The Matrix'", 'role': 'assistant', 'name': 'user_proxy'}, {'content': "Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving are a few actors who've played in 'The Matrix'.", 'role': 'user', 'name': 'matrix_agent'}, {'content': 'Any additional actors?', 'role': 'assistant', 'name': 'user_proxy'}, {'content': "I'm sorry, but I don't have information about any additional actors from 'The Matrix' beyond Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving.", 'role': 'user', 'name': 'matrix_agent'}], summary="I'm sorry, but I don't have information about any additional actors from 'The Matrix' beyond Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving.", cost={'usage_including_cached_inference': {'total_cost': 0}, 'usage_excluding_cached_inference': {'total_cost': 0}}, human_input=['Any additional actors?', 'exit'])

Revisit the example with self defined graph ontology

If you want to be more specific on the entities and attributes in the graph database you can create an Ontology before loading your data, in the case below we define these entities (Actor with name and Movie with title) and relationships (Actors ACTED in Movies). This allows the RAG agent to answer questions about actors in the movie.

from graphrag_sdk import Attribute, AttributeType, Entity, Ontology, Relation

from autogen.agentchat.contrib.graph_rag.document import Document, DocumentType
from autogen.agentchat.contrib.graph_rag.falkor_graph_query_engine import FalkorGraphQueryEngine

input_path = "../test/agentchat/contrib/graph_rag/the_matrix.txt"

movie_ontology = Ontology()
movie_ontology.add_entity(
    Entity(label="Actor", attributes=[Attribute(name="name", attr_type=AttributeType.STRING, unique=True)])
)
movie_ontology.add_entity(
    Entity(label="Movie", attributes=[Attribute(name="title", attr_type=AttributeType.STRING, unique=True)])
)
movie_ontology.add_relation(Relation(label="ACTED", source="Actor", target="Movie"))

query_engine = FalkorGraphQueryEngine(
    name="IMDB",
    host="172.17.0.4",  # Change
    port=6379,  # if needed
    ontology=movie_ontology,
)

input_documents = [Document(doctype=DocumentType.TEXT, path_or_url=input_path)]

query_engine.init_db(input_doc=input_documents)

Add capability to a ConversableAgent and query them

With FalkorDB setup with an ontology and data, we can now associate the GraphRAG capability with a ConversableAgent and have a chat with it.

from autogen import ConversableAgent, UserProxyAgent
from autogen.agentchat.contrib.graph_rag.falkor_graph_rag_capability import FalkorGraphRagCapability

# Create a ConversableAgent (no LLM configuration)
graph_rag_agent = ConversableAgent(
    name="matrix_agent",
    human_input_mode="NEVER",
)

# Associate the capability with the agent
graph_rag_capability = FalkorGraphRagCapability(query_engine)
graph_rag_capability.add_to_agent(graph_rag_agent)

# Create a user proxy agent to converse with our RAG agent
user_proxy = UserProxyAgent(
    name="user_proxy",
    code_execution_config=False,
    is_termination_msg=lambda msg: "TERMINATE" in msg["content"],
    human_input_mode="ALWAYS",
)

user_proxy.initiate_chat(graph_rag_agent, message="Name a few actors who've played in 'The Matrix'")

# You will be prompted, as a human in the loop, after the response - feel free to ask more questions.

user_proxy (to matrix_agent):

Name a few actors who've played in 'The Matrix'

--------------------------------------------------------------------------------
matrix_agent (to user_proxy):

Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving have played in 'The Matrix'.

--------------------------------------------------------------------------------
user_proxy (to matrix_agent):

Who else acted in The Matrix?

--------------------------------------------------------------------------------
matrix_agent (to user_proxy):

Based on the provided information, Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving are the actors listed for 'The Matrix'. No additional actors are mentioned in the data.

--------------------------------------------------------------------------------
user_proxy (to matrix_agent):

Where was The Matrix filmed?

--------------------------------------------------------------------------------
matrix_agent (to user_proxy):

I am sorry, I could not find the answer to your question

--------------------------------------------------------------------------------

ChatResult(chat_id=None, chat_history=[{'content': "Name a few actors who've played in 'The Matrix'", 'role': 'assistant', 'name': 'user_proxy'}, {'content': "Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving have played in 'The Matrix'.", 'role': 'user', 'name': 'matrix_agent'}, {'content': 'Who else acted in The Matrix?', 'role': 'assistant', 'name': 'user_proxy'}, {'content': "Based on the provided information, Keanu Reeves, Laurence Fishburne, Carrie-Anne Moss, and Hugo Weaving are the actors listed for 'The Matrix'. No additional actors are mentioned in the data.", 'role': 'user', 'name': 'matrix_agent'}, {'content': 'Where was The Matrix filmed?', 'role': 'assistant', 'name': 'user_proxy'}, {'content': 'I am sorry, I could not find the answer to your question', 'role': 'user', 'name': 'matrix_agent'}], summary='I am sorry, I could not find the answer to your question', cost={'usage_including_cached_inference': {'total_cost': 0}, 'usage_excluding_cached_inference': {'total_cost': 0}}, human_input=['Who else acted in The Matrix?', 'Where was The Matrix filmed?', 'exit'])

From the output we can see that: - We can ask about actors and movies because we defined the ontology (and they were also automatically created in the prior example) - There is inbuilt awareness of answers it has provided previously - It can’t answer any questions outside of the actors and movies entities as we didn’t include anything else in the ontology

# Remove graph from database:
query_engine.delete()

True