06. Agentic RAG

In this chapter, we will create an agent that accesses the latest information through document search and generates answers with search results.

Let's create an agent that searches for and answers documents based on your questions, or uses Internet search tools to answer them.

Reference

• If you do RAG, but use Agent to do RAG, Agentic RAG It is called.

Tools

You can define the tools that Agent will utilize to make them utilized when performing Agent's inference.

Tavily Search is a representative of them Search tools is. Search allows you to access the latest information and generate answers with search results. Tools offer a variety of different types and methodologies, including search tools like this, tools that can run Python code, and tools that run directly defined functions.

Web Search Tool: Tavily Search

LangChain has built-in tools that make Tavily search engines easy to use as tools.

API KEY must be issued to use Tavily Search.

• Tavily Search API issued

Register the issued API KEY for environmental variables as follows:

.env Register as follows in the file:

• TAVILY_API_KEY=Enter the issued Tavily API KEY

Copy

# Configuration file for managing API keys as environment variables
from dotenv import load_dotenv

# Load API key infomation
load_dotenv()

Copy

 True 

Copy

# LangSmith set up tracking. https://smith.langchain.com
# !pip install -qU langchain-teddynote
from langchain_teddynote import logging

# Enter a project name.
logging.langsmith("CH15-Agents")

Copy

 Start tracking LangSmith. 
[Project name] 
CH15-Agents 

Copy

# TavilySearchResults the class langchain_community.tools.tavily_search 모듈에서 가져옵니다.
from langchain_community.tools.tavily_search import TavilySearchResults

# Create an instance of the TavilySearchResults class.
# k=6 means that we will retrieve up to 6 search results.
search = TavilySearchResults(k=6)

search.invoke The function executes a search for a given string.

invoke() Perform a search by putting the search terms you want to search for in the function.

Copy

# Get search results.
search.invoke("What is the phone number of the Pangyo Kakao Friends Shop Ajit branch?")

Copy

 [{'url':'https://blog.naver.com/PostView.naver?blogId=amour1115&logNo=223386213560','content':'Chao Friends Pansy Azit. . Closed all year round 12:00-21:00. Sat, day 12:00-20:00. . 166. Kakao Pansa Azit 1st floor Kakao Friends. 031-301-7225. This is a non-existent image.'}, {'url':'https://kr.trip.com/moments/detail/seongnam-121856-121058702/','content':'If you want to enjoy indoor dating while taking pictures, pay attention 📢 Baby girl to watch at the Pantagi Cute Hot Place Kakao Friends of the characters 
(...sway...) 
🔖Bathing hot in the in-star vernalgai だ\n Now Kakhao Franz's great deal is vernal.\n 🔖Sojudisian Jordi 39,000 won\n🔖 Electric and Inwoo Knuts \n23,000 won\n🔖 Riklimb💡\n Cara Spring is a liqueur!!!\n'}] 

Document-based document search tool: Retriever

It also generates retrievers to perform queries on our data.

Documents utilized for practice

Software Policy Institute (SPRi)-December 2023

• Author: Jaeheung Lee (AI Policy Institute Office Liability Institute), Lee Ji-soo (AI Policy Lab Yi Phyang Institute)

• Link: https://spri.kr/posts/view/23669

• File name: SPRI_AI_Brief_2023년12월호_F.pdf

Files downloaded for practice data Please copy to folder

This code builds a document retrieval system using web-based document loaders, document splitters, vector repositories, and OpenAI embedding.

PDF documents here FAISS Generates retrievers that are stored and viewed in DB.

Copy

from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings
from langchain.document_loaders import PyPDFLoader

# Load PDF file. Enter path to file.
loader = PyPDFLoader("data/SPRI_AI_Brief_2023년12월호_F.pdf")

# Split your document using the text splitter.
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)

# Load and split the document.
split_docs = loader.load_and_split(text_splitter)

# Create a VectorStore.
vector = FAISS.from_documents(split_docs, OpenAIEmbeddings())

# Create a Retriever.
retriever = vector.as_retriever()

This function retriever Object invoke() Use to most of your questions Relevance documents Used to find.

Copy

# Get relevant documents from your document.
retriever.invoke("Please find the information related to generative AI developed by Samsung Electronics in the document.")

Copy

 [Document (metadata={'source':'data/SPRI_AI_Brief_2023 Year 12 _F.pdf','page': 12}'SPRi AI Brief = \n2023-12 month The \n Samsung Gauss, capable of on-device operation, has the advantage that there is no risk of user information leaking out of the KEY Contents\n£ Samsung Gauss, consisting of three models of language, code, and images, 
(...sway...) 
If you're staying, the agent analyzes the user's travel \npaton to suggest a destination, recommend activities based on interests, and book a restaurant in your preferred style \n£AI agent will lead the popularization of services in the medical and educational, productivity, entertainment and shopping areas \ Nn A notable change due to the transfer of △Medical △Productivity <TA 

Now that we have filled the index to perform the search, we can easily convert it into a tool that agents can use properly.

create_retriever_tool As a function retriever Convert to tools.

Copy

from langchain.tools.retriever import create_retriever_tool

retriever_tool = create_retriever_tool(
    retriever,
    name="pdf_search",  # Enter a name for the tool.
    description="use this tool to search information from the PDF document",  # 도구에 대한 설명을 자세히 기입해야 합니다!!
)

Agent Defines a list of tools to use

Now that you have created both, you can create a list of tools for Agent to use.

tools List search Wow retriever_tool Includes.

Copy

# Add search and retriever_tool to the tools list.
tools = [search, retriever_tool]

Agent creation

Now that you have defined the tool, you can create an agent.

First, the Agent defines the LLM to utilize, and the Agent defines the Prompt to refer to.

Reference -If you don't support multi-turn conversations, you can remove "chat_history".

Copy

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate

# LLM 정의
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)

# Prompt 정의
prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are a helpful assistant. "
            "Make sure to use the `pdf_search` tool for searching information from the PDF document. "
            "If you can't find the information from the PDF document, use the `search` tool for searching information from the web.",
        ),
        ("placeholder", "{chat_history}"),
        ("human", "{input}"),
        ("placeholder", "{agent_scratchpad}"),
    ]
)

Next, create a Tool Calling Agent.

Copy

from langchain.agents import create_tool_calling_agent

# tool calling agent generation
agent = create_tool_calling_agent(llm, tools, prompt)

Finally, created agent Run AgentExecutor Generate.

Reference

• verbose=False Set to omit the intermediate step output.

Copy

from langchain.agents import AgentExecutor

# AgentExecutor generation
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=False)

Execute agent

Now you can run the agent for some queries!

All these queries are now Stateless It is a question (I do not remember the previous interaction).

agent_executor Object invoke Methods are processed by taking a factor in the form of a dictionary. In this example input Kier hi! I am passing the dictator assigned a value as a factor. It is commonly used to process input from objects such as AI agents, function launchers, or command handlers.

Copy

from langchain_teddynote.messages import AgentStreamParser

# Generate parsers for each step output
agent_stream_parser = AgentStreamParser()

Copy

# Request streaming output of answers to queries
result = agent_executor.stream(
    {"input": "Please search and let me know the 5 teams that advanced to the 2024 professional baseball playoffs."}
)

for step in result:
    # Output the intermediate steps step by step using parser
    agent_stream_parser.process_agent_steps(step)

Copy

 [Tool call] 
Tool: tavily_search_results_json 
query: 2024 year pro baseball playoff advancement team 
Log:  
Invoking: `tavily_search_results_json` with `{'query': '2024 Pro Baseball Playoff Advance Team' }` 



[Observation content] 
Observation: [{'url':'https://www.yna.co.kr/view/GYH20241003000600044','content': '4 days ago... The team's flipping magic was preempted and advanced to the semi-playoff (3rd round before JPO·5). kt was held at the 2024 New SOLbank KBO Post Season Wild\xa0... '}, 
(...sway...) 
{'url':'https://namu.wiki/w/KBO%20%ED%94%8C%EB%A0%88%EC%9D%B4%EC%98%A4%ED%94%84/2024%EB%85%84','content': '2024 The series of new SOL Bank KBO League Pennantrace 2nd Samsung Lions and the Junplayoff Victory Team.'}] 

[Tool call] 
Tool: tavily_search_results_json 
query: 2024 KBO postseason team 
Log:  
Invoking: `tavily_search_results_json` with `{'query': '2024 KBO Post Season Advance Team'}` 

[Observation content] 
Observation: [{'url':'https://www.yna.co.kr/view/GYH20241003000600044','content': '4 days ago... The team's flipping magic was preempted and advanced to the semi-playoff (3rd round before JPO·5). kt was held at the 2024 New SOLbank KBO Post Season Wild\xa0... '},  
(...sway...) 
{'url':'http://m.monthly.chosun.com/client/Mdaily/daily_view.asp?Idx=20426&Newsnumb=20241020426','content': '5 days ago · The advance team is ▲KIA Tigers ▲Samsung Lions ▲LG Twins ▲Doosan Bears ▲kt wiz. On the 2nd, Doosan and KT play wildcard crystals in Jamsil-gu, Seoul.'}] 

[Final answer] 
The teams that entered the 2024 KBO Pro Baseball Postseason are: 

One. KIA Tigers 
2. Samsung Lions 
3. LG Twins 
4. Doosan Bears 
5. KT Wiz 

These teams will compete in the postseason. 

agent_executor Object invoke Using the method, the question is provided as input.

Copy

# Request streaming output of answers to queries
result = agent_executor.stream(
    {"input": "Please find information related to the generative AI developed by Samsung Electronics in the document."}
)

for step in result:
    # Output the intermediate steps step by step using parser
    agent_stream_parser.process_agent_steps(step)

Copy

 [Tool call] 
Tool: pdf_search 
query: Samsung-generated AI 
Log:  
Invoking: `pdf_search` with `{'query':'Triple-generated AI'}` 

[Observation content] 
Observation: SPRi AI Brief |  
December 2023 
10 Samsung Electronics unveils self-developed AI ‘Samsung Gauss ’ 
Self-developed creation with 3 models of language, code, and images, with n-three-scriptors working on the on-divide  
AI model ‘Samsung Gauss ’released 
(...sway...) 
The n-judgment is also copyrighted on all images used to train the generated AI model, or made with the generated AI  
Since the image was trained using the work, the claim that it is a derivative image of the work lacks personality.  
Point out 

[Final answer] 
Here's information on the Samsung-developed AI model'Samsung Gauss': 

One. **Model Configuration**: Samsung Gauss consists of 3 models of language, code and images.  
   -**Language model**: Supports mail writing, document summarization, translation work, and more. 
   -**Code Model**: AI coding assistant'code.i' provides services with an interactive interface and is optimized for in-house software development. 
   -**Image model**: You can create creative images, modify existing images, and also support the ability to convert low resolution images to high resolution. 

2. **Ondice works**: Samsung Gauss is designed to work on the Ondevice, so there is no risk of user information leaking outward. This serves as a great advantage in terms of privacy. 

3. **Release and Planning**: The Samsung Electronics first unveiled Samsung Gauss at the'Samsung AI Forum 2023' on November 8, 2023, and plans to phase out various products. 

4. **Technology Trends**: According to IT Specialization, Ondice AI is emerging as a major technology trend, and since 2024, smartphones with Samsung Gauss will compete with Meta's Lama2-equipped quality-com devices and Google-assisted Google Pixel It is expected. 

This information demonstrates Samsung's efforts to enhance the user experience through AI technology and enhance privacy. 

Agent to remember the previous conversation

To remember the previous conversation RunnableWithMessageHistory Using AgentExecutor Wraps.

RunnableWithMessageHistory Please refer to the link below for more information on this.

Reference

• RunnableWithMessageHistory

Copy

from langchain_community.chat_message_histories import ChatMessageHistory
from langchain_core.runnables.history import RunnableWithMessageHistory

# Create a dictionary to store session_id
store = {}


# Function to retrieve session records based on session_id
def get_session_history(session_ids):
    if session_ids not in store:  # If session_id is not in the store
        # Create a new ChatMessageHistory object and save it to the store.
        store[session_ids] = ChatMessageHistory()
    return store[session_ids]  # Returns the session record for the given session ID.


# Create an agent with added chat message history.
agent_with_chat_history = RunnableWithMessageHistory(
    agent_executor,
    # Conversation session_id
    get_session_history,
    # Key: "input" where the question in the prompt is entered
    input_messages_key="input",
    # The key that the prompt's message is entered into: "chat_history"
    history_messages_key="chat_history",
)

Copy

# Request streaming output of answers to queries
response = agent_with_chat_history.stream(
    {"input": "Please find information related to generative AI developed by Samsung Electronics in the document."},
    # session_id 설정
    config={"configurable": {"session_id": "abc123"}},
)

# Check the output
for step in response:
    agent_stream_parser.process_agent_steps(step)

Copy

 [Tool call] 
Tool: pdf_search 
query: Samsung-generated AI 
Log:  
Invoking: `pdf_search` with `{'query':'Triple-generated AI'}` 

[Observation content] 
Observation: SPRi AI Brief |  
December 2023 
10 Samsung Electronics unveils self-developed AI ‘Samsung Gauss ’ 
Self-developed creation with 3 models of language, code, and images, with n-three-scriptors working on the on-divide  
AI model ‘Samsung Gauss ’released 
(...sway...) 
∙The ruling is copyrighted by the US Copyright Office for the work of art by Callie McCannan and Carla Ortiz, who filed the lawsuit.  
Pointed out that it was not submitted, and Sarah Anderson wrote 16 of the hundreds cited in the complaint.  
Copyright only 
The n-judgment is also copyrighted on all images used to train the generated AI model, or made with the generated AI  
Since the image was trained using the work, the claim that it is a derivative image of the work lacks personality.  
Point out 

[Final answer] 
Here is information on the generated AI developed by the Samsung: 

One. **Samsung Gauss**: The Samsung Electronics unveiled its own developed AI model,'Samsung Gauss', at the'Samsung AI Forum 2023' on November 8, 2023. This model consists of 3 models of language, code, and images, and is designed to work on an on-device. This has the advantage that there is no risk of user information leaking outward. 

2. **Model configuration**: 
   -**Language model**: Consists of various models in the cloud and on-the-fly, supporting tasks such as composing mail, summarizing documents, and translating. 
   -**Code Model**: Provides services with an interactive interface through the AI coding assistant'code.i' and is optimized for in-house software development. 
   -**Image model**: You can create creative images, modify existing images, and also support the ability to convert low resolution images to high resolution. 

3. **Future Plan**: The Samsung Electronics plans to phase out Samsung Gauss in a variety of products, and introduces several features utilizing on-the-case AI technology. The technology will be on Samsung smartphones starting in 2024, and is expected to compete with Meta's Llama2-equalcom device and Google Pixel with Google Assistant. 

This information demonstrates Samsung's efforts to enhance the user experience and enhance data security through AI technology. 

Copy

response = agent_with_chat_history.stream(
    {"input": "Please translate the previous answer into English."},
    # Set session_id
    config={"configurable": {"session_id": "abc123"}},
)

# Check the output
for step in response:
    agent_stream_parser.process_agent_steps(step)

Copy

 [Final answer] 
Here is the translation of the previous response into English: 

One. **Samsung Gauss**: On November 8, 2023, Samsung Electronics unveiled its generative AI model called'Samsung Gauss' at the'Samsung AI Forum 2023'. This model consists of three components: language, code, and image models, and is designed to operate on-device. This has the advantage of ensuring that user information is not at risk of being leaked externally. 

2. **Model Composition**: 
   -**Language Model**: Composed of various models for cloud and on-device use, it supports tasks such as email writing, document summarization, and translation. 
   -**Code Model**: Through the AI coding assistant'code.i', it provides services via an interactive interface, optimized for in-house software development. 
   -**Image Model**: Capable of generating creative images and modeling existing ones, it also supports permission low-resolution images to high-resolution. 

3. **Future Plans**: Samsung Electronics plans to gradually integrate Samsung Gauss into various products and introduce multiple features utilizing on-device AI technology. This technology is excellent to be incorporated into Samsung smartphones starting in 2024 and is anticipated to compete with Qualcomm devices equipped with Meta's Llama 2 and Google Pixel devices using Google Assistant. 

This information demonstrates Samsung's effics to enhance user experience and strengthen data security through AI technology. 

Agent template

Here is the full template code.

Copy

# Import required modules
from langchain_core.prompts import ChatPromptTemplate
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_community.vectorstores import FAISS
from langchain_community.chat_message_histories import ChatMessageHistory
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain.document_loaders import PyMuPDFLoader
from langchain.tools.retriever import create_retriever_tool
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_teddynote.messages import AgentStreamParser

########## 1. Define a tool ##########

### 1-1. Search Tool ###
# Create an instance of the TavilySearchResults class.
# k=6 means that we will retrieve up to 6 search results.
search = TavilySearchResults(k=6)

### 1-2. PDF Document Search Tool (Retriever) ###
# Load PDF file. Enter path to file.
loader = PyMuPDFLoader("data/SPRI_AI_Brief_2023년12월호_F.pdf")

# Split your document using the text splitter.
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)

# Load and split the document.
split_docs = loader.load_and_split(text_splitter)

# Create a VectorStore.
vector = FAISS.from_documents(split_docs, OpenAIEmbeddings())

# Create a Retriever.
retriever = vector.as_retriever()

retriever_tool = create_retriever_tool(
    retriever,
    name="pdf_search",  # Enter a name for the tool.
    description="use this tool to search information from the PDF document",  # 도구에 대한 설명을 자세히 기입해야 합니다!!
)

### 1-3. Add a list of tools to the tools list ###
# Add search and retriever_tool to the tools list.
tools = [search, retriever_tool]

########## 2. Define LLM ##########
# Create an LLM model.
llm = ChatOpenAI(model="gpt-4o", temperature=0)

########## 3. Define Prompt ##########

# Define the prompt - you can modify this part!
# Prompt definition

prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are a helpful assistant. "
            "Make sure to use the `pdf_search` tool for searching information from the PDF document. "
            "If you can't find the information from the PDF document, use the `search` tool for searching information from the web.",
        ),
        ("placeholder", "{chat_history}"),
        ("human", "{input}"),
        ("placeholder", "{agent_scratchpad}"),
    ]
)

########## 4. Define Agent ##########

# Create an agent.
# llm, tools, It takes prompt as an argument.
agent = create_tool_calling_agent(llm, tools, prompt)

########## 5. Define AgentExecutor ##########

# Set up the agent and tools using the AgentExecutor class, and set verbose to True to output detailed logs.
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=False)

########## 6. Add memory to perform chat logging. ##########

# Create a dictionary to store session_id
store = {}


# Function to retrieve session records based on session_id
def get_session_history(session_ids):
    if session_ids not in store:  # If session_id is not in the store
        # Create a new ChatMessageHistory object and save it to the store.
        store[session_ids] = ChatMessageHistory()
    return store[session_ids]  # Returns the session record for the given session ID.


# Create an agent with added chat message history.
agent_with_chat_history = RunnableWithMessageHistory(
    agent_executor,
    # Conversation session_id
    get_session_history,
    # Key: "input" where the question in the prompt is entered
    input_messages_key="input",
    # The key that the prompt's message is entered into: "chat_history"
    history_messages_key="chat_history",
)

########## 7. Define an agent parser ##########
agent_stream_parser = AgentStreamParser()

Copy

########## 8. Run the agent and check the results. ##########

# Prints answers to queries.
response = agent_with_chat_history.stream(
    {"input": "Find the amount Google invested in Anthropic in the document."},
    # Sets the session ID.
    # This is not actually used here since we use a simple in-memory ChatMessageHistory
    config={"configurable": {"session_id": "abc123"}},
)

for step in response:
    agent_stream_parser.process_agent_steps(step)

Copy

 [Tool call] 
Tool: pdf_search 
query: Google and Pacific investment amount 
Log:  
Invoking: `pdf_search` with `{'query':'Google and Pacific Investment Amount' }` 

[Observation content] 
Observation: 1. Policy/legal  
2. Enterprise/Industry  
3. Technology / Research  
 4. Workforce/training 
Google strengthens AI cooperation generated by $20 billion investment in Aspic  
n Google has agreed to invest up to $20 billion in Asropic and $500 million in priority, and Ansropic  
Google also signed a contract to use cloud services. 
(...sway...)  
Ⅱ. Main event 
   ▹CES 2024 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·· · · · ·· · · ·· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ······· · · · · · · · · · ·· · · · · · · · · · · · · · · · · · · · · · · · · · · · 

[Final answer] 
Google has agreed to invest up to $200 billion in Ansropic, of which $500 million has been invested first. We plan to invest an additional $1.5 billion in the future. 

Copy

########## 8. Run the agent and check the results. ##########

# Prints answers to queries.
response = agent_with_chat_history.stream(
    {"input": "Please translate the previous answer into English"},
    # Sets the session ID.
    # This is not actually used here since we use a simple in-memory ChatMessageHistory
    config={"configurable": {"session_id": "abc123"}},
)

for step in response:
    agent_stream_parser.process_agent_steps(step)

Copy

 [Final answer] 
Google has agreed to invest up to $2 billion in Anthropic, with an initial investment of $500 million. They plan to invest an additional $1.5 billion in the future. 

Copy

########## 8. Run the agent and check the results. ##########

# Prints answers to queries.
response = agent_with_chat_history.stream(
    {
        "input": "Please search and tell me the 5 teams that will advance to the 2024 professional baseball playoffs. Please answer in Korean."
    },
    # Sets the session ID.
    # This is not actually used here since we use a simple in-memory ChatMessageHistory
    config={"configurable": {"session_id": "abc456"}},
)

for step in response:
    agent_stream_parser.process_agent_steps(step)

Copy

 [Tool call] 
Tool: tavily_search_results_json 
query: 2024 KBO League Playoff Advance Team 
Log:  
Invoking: `tavily_search_results_json` with `{'query': '2024 KBO League Playoff Advance Team'}` 

[Observation content] 
Observation: [{'url':'https://www.gosusports.com/news/baseball/kbo-news-2024-kbo-season-preview/196180/','content': "[High Sports] 2024 kbo League Season Preview... Playoff advancement. In nc's propaganda, the role of'special outsider' pedi was in effect. Pedi was all ranked 1st in the trivalent sector, including the average bookstore (era 2.00), 20 wins, and threesome (209), making kborigs fair."}, 
(...sway...) 
kia Tigers, Samsung Lions, lg Twins, Doosan Bears, kt Wizda.. kia Tigers; After 2022, he succeeded in entering the postseason. After the confirmation of the postseason entry on September 6, the regular season championship on September 17 was also successful in the Korean series.'}] 

[Final answer] 
The five teams that entered the KBO League playoffs in 2024 are: 

One. KIA Tigers 
2. Samsung Lions 
3. LG Twins 
4. Doosan Bears 
5. KT Wiz 

These teams will advance to the KBO postseason in 2024 and play. 

Copy

########## 8. Run the agent and check the results. ##########

# Prints answers to queries.
response = agent_with_chat_history.stream(
    {"input": "Please write your previous answer in the form of a SNS post, within 100 characters."},
    # Sets the session ID.
    # It's not really used here because we use a simple in-memory ChatMessageHistory
    config={"configurable": {"session_id": "abc456"}},
)

for step in response:
    agent_stream_parser.process_agent_steps(step)

Copy

 [Final answer] 
2024 KBO League Playoff Advance Team: KIA Tigers, Samsung Lions, LG Twins, Doosan Bears, KT Wiz! Fierce matches are expected! #KBO # Playoff 

Copy

########## 8. Run the agent and check the results. ##########

# Prints answers to queries.
response = agent_with_chat_history.stream(
    {"input": "Please add the Korean Series schedule to the previous answer."},
    # Sets the session ID.
    # It's not really used here because we use a simple in-memory ChatMessageHistory
    config={"configurable": {"session_id": "abc456"}},
)

for step in response:
    agent_stream_parser.process_agent_steps(step)

Copy

 [Tool call] 
Tool: tavily_search_results_json 
query: 2024 KBO Korea Series Schedule 
Log:  
Invoking: `tavily_search_results_json` with `{'query': '2024 KBO Korea Series Schedule'}` 

[Observation content] 
Observation: [{'url':'https://ko.wikipedia.org/wiki/2024_Korea Series','content': '2024 The new SOL Bank KBO Korea Series is the champion decision of the 2024 KBO League, which is scheduled to take place from October 21 to October 29, 2024. Dajin is the No. 1 team in the regular season of the KBO League in 2024, with KIA Tigers who succeed in directing the Korean Series in 7 years after the Korean Series in 2017 and challenge the unified championship...'},  
(...sway...) 
Can be shipped. On the other hand, the fifth-place team must win both matches to get to the semi-playoff. If the match ends in a draw, the match will be played in this series without a moving date in the ballpark where the draw occurred after the final match by series. WIKITREE WIKITREE NOW" }] 

[Final answer] 
2024 KBO League Playoff Advance Team: KIA Tigers, Samsung Lions, LG Twins, Doosan Bears, KT Wiz! The match starts on October 2nd, and the Korean series runs from October 21st to October 29th. Fierce matches are expected! #KBO # Playoff # Korea Series