vinagent 0.0.1

No project description provided

1. Introduction

agentcore is a lightweight and flexible library designed for building smart agent assistants across various industries. Whether you're creating an AI-powered customer service bot, a data analysis assistant, or a domain-specific automation agent, agentcore provides a simple yet powerful foundation.

With its modular tool system, you can easily extend your agent's capabilities by integrating a wide range of tools. Each tool is self-contained, well-documented, and can be registered dynamically—making it effortless to scale and adapt your agent to new tasks or environments.

To install and use this library please following:

[e-mail]:datascienceworld-kan/agentcore.git
cd agentcore
pip install -r requirements.txt
poetry install

To use a list of default tools inside agentcore.tools you should set environment varibles inside .env including TOGETHER_API_KEY to use llm models at togetherai site and TAVILY_API_KEY to use tavily websearch tool at tavily site:

TOGETHER_API_KEY="Your together API key"
TAVILY_API_KEY="Your Tavily API key"

Let's create your acounts first and then create your relevant key for each website.

2. Set up Agent

agentcore is a flexible library for creating intelligent agents. You can configure your agent with tools, each encapsulated in a Python module under agentcore.tools. This provides a workspace of tools that agents can use to interact with and operate in the realistic world. Each tool is a Python file with full documentation and it can be independently ran. For example, the agentcore.tools.websearch_tools module contains code for interacting with a search API.

from langchain_together import ChatTogether 
from agentcore.agent.agent import Agent
from dotenv import load_dotenv
load_dotenv()

llm = ChatTogether(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
)

# Step 1: Create Agent with tools
agent = Agent(
    description="You are a Financial Analyst",
    llm = llm,
    skills = [
        "Deeply analyzing financial markets", 
        "Searching information about stock price",
        "Visualization about stock price"],
    tools = ['agentcore.tools.websearch_tools',
             'agentcore.tools.yfinance_tools'],
    tools_path = 'templates/tools.json' # Place to save tools. Default is 'templates/tools.json',
    is_reset_tools = True # If True, will reset tools every time. Default is False
)

# Step 2: invoke the agent
message = agent.invoke("Who you are?")

If the answer is a normal message without using any tools, it will be an AIMessage. By contrast, it will have ToolMessage type. For examples:

message

AIMessage(content='I am a Financial Analyst.', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 7, 'prompt_tokens': 308, 'total_tokens': 315, 'completion_tokens_details': None, 'prompt_tokens_details': None, 'cached_tokens': 0}, 'model_name': 'meta-llama/Llama-3.3-70B-Instruct-Turbo-Free', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-070f7431-7176-42a8-ab47-ed83657c9463-0', usage_metadata={'input_tokens': 308, 'output_tokens': 7, 'total_tokens': 315, 'input_token_details': {}, 'output_token_details': {}})

Access to content property to get the string content.

message.content

I am a Financial Analyst.

The following function need to use yfinancial tool, therefore the return value will be ToolMessage with the a stored pandas.DataFrame in artifact property.

df = agent.invoke("What is the price of Tesla stock in 2024?")
df

ToolMessage(content="Completed executing tool fetch_stock_data({'symbol': 'TSLA', 'start_date': '2024-01-01', 'end_date': '2024-12-31', 'interval': '1d'})", tool_call_id='tool_cde0b895-260a-468f-ac01-7efdde19ccb7', artifact=pandas.DataFrame)

To access pandas.DataFrame value:

df.artifact.head()

Another example, if you visualize a stock price using a tool, the output message is a ToolMessage with the saved artifact is a plotly plot.

# return a ToolMessage which we can access to plot by plot.artifact and content by plot.content.
plot = agent.invoke("Let's visualize Tesla stock in 2024?")

# return a ToolMessage which we can access to plot by plot.artifact and content by plot.content.
plot = agent.invoke("Let's visualize the return of Tesla stock in 2024?")

3. Register function tool

Function tools are registered directly in your runtime code by decorating them with the @function_tool without saving them into python module files.

from agentcore.register.tool import function_tool
from typing import List

@function_tool
def sum_of_series(x: List[float]):
    return f"Sum of list is {sum(x)}"

INFO:root:Registered tool: sum_of_series (runtime)

message = agent.invoke("Sum of this list: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]?")
message

ToolMessage(content="Completed executing tool sum_of_series({'x': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]})", tool_call_id='tool_56f40902-33dc-45c6-83a7-27a96589d528', artifact='Sum of list is 55')

4. Advance Features

4.1. Deep Search

With agentcore, you can invent a complex workflow by combining multiple tools into a single agent. This allows you to create a more sophisticated and flexible agent that can adapt to different task. Let's see how an agent can be created to help with financial analysis by using deepsearch tool, which allows you to search for information in a structured manner. This tool is particularly useful for tasks that require a deep understanding of the data and the ability to navigate through complex information.

from langchain_together import ChatTogether 
from agentcore.agent import Agent
from dotenv import load_dotenv
load_dotenv()

llm = ChatTogether(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
)

agent = Agent(
    description="You are a Financial Analyst",
    llm = llm,
    skills = [
        "Deeply analyzing financial markets", 
        "Searching information about stock price",
        "Visualization about stock price"],
    tools = ['agentcore.tools.deepsearch']
)
    
message = agent.invoke("Let's analyze Tesla stock in 2025?")
print(message.artifact)

The output is available at agentcore/examples/deepsearch.md

4.2. Trending Search

Exceptionally, agentcore also offers a feature to summarize and highlight the top daily news on the internet based on any topic you are looking for, regardless of the language used. This is achieved by using the trending_news tool.

from langchain_together import ChatTogether 
from agentcore.agent.agent import Agent
from dotenv import load_dotenv
load_dotenv()

llm = ChatTogether(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
)

agent = Agent(
    description="You are a Trending News Analyst",
    llm = llm,
    skills = [
        "Searching the trending news on realtime from google news",
        "Deeply analyzing top trending news"],
    tools = ['agentcore.tools.trending_news']
)
    
message = agent.invoke("Tìm 5 tin tức nổi bật về tình hình giá vàng sáng hôm nay")
print(message.artifact)

The output is available at agentcore/examples/todaytrend.md

5. Agent with Memory (>=0.1.2)

There is a special upgradation from version 0.1.2 that allows to adhere Memory for each Agent. This is useful when you want to keep track of the user's behavior and conceptualize them as a knowledge graph for the agent. Therefore, it helps agent become more intelligent and capable of understanding personality and responding to user queries with greater accuracy.

The following code to save each conversation into short-memory.

from langchain_together.chat_models import ChatTogether
from dotenv import load_dotenv
from agentcore.memory import Memory

load_dotenv()

llm = ChatTogether(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
)

memory = Memory(
    memory_path="templates/memory.jsonl",
    is_reset_memory=True, # will reset the memory every time the agent is invoked
    is_logging=True
)

text_input = """Hi, my name is Kan. I was born in Thanh Hoa Province, Vietnam, in 1993.
My motto is: "Make the world better with data and models". That’s why I work as an AI Solution Architect at FPT Software and as an AI lecturer at NEU.
I began my journey as a gifted student in Mathematics at the High School for Gifted Students, VNU University, where I developed a deep passion for Math and Science.
Later, I earned an Excellent Bachelor's Degree in Applied Mathematical Economics from NEU University in 2015. During my time there, I became the first student from the Math Department to win a bronze medal at the National Math Olympiad.
I have been working as an AI Solution Architect at FPT Software since 2021.
I have been teaching AI and ML courses at NEU university since 2022.
I have conducted extensive research on Reliable AI, Generative AI, and Knowledge Graphs at FPT AIC.
I was one of the first individuals in Vietnam to win a paper award on the topic of Generative AI and LLMs at the Nvidia GTC Global Conference 2025 in San Jose, USA.
I am the founder of DataScienceWorld.Kan, an AI learning hub offering high-standard AI/ML courses such as Build Generative AI Applications and MLOps – Machine Learning in Production, designed for anyone pursuing a career as an AI/ML engineer.
Since 2024, I have participated in Google GDSC and Google I/O as a guest speaker and AI/ML coach for dedicated AI startups.
"""

memory.save_short_term_memory(llm, text_input)
memory_message = memory.load_memory('string')
memory_message

Kan -> BORN_IN[in 1993] -> Thanh Hoa Province, Vietnam
Kan -> WORKS_FOR[since 2021] -> FPT Software
Kan -> WORKS_FOR[since 2022] -> NEU
Kan -> STUDIED_AT -> High School for Gifted Students, VNU University
Kan -> STUDIED_AT[graduated in 2015] -> NEU University
Kan -> RESEARCHED_AT -> FPT AIC
Kan -> RECEIVED_AWARD[at Nvidia GTC Global Conference 2025] -> paper award on Generative AI and LLMs
Kan -> FOUNDED -> DataScienceWorld.Kan
Kan -> PARTICIPATED_IN[since 2024] -> Google GDSC
Kan -> PARTICIPATED_IN[since 2024] -> Google I/O
DataScienceWorld.Kan -> OFFERS -> Build Generative AI Applications
DataScienceWorld.Kan -> OFFERS -> MLOps – Machine Learning in Production
Kan -> OWNS -> house
Kan -> OWNS -> garden
Kan -> HAS_MOTTO -> stay hungry and stay foolish
Kan -> RECEIVED_AWARD_AT[in 2025] -> Nvidia GTC Global Conference

To adhere Memmory to each Agent

import os
import sys
from langchain_together import ChatTogether 
from agentcore.agent import Agent
from agentcore.memory.memory import Memory
from pathlib import Path
from dotenv import load_dotenv
load_dotenv()

llm = ChatTogether(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
)

# Step 1: Create Agent with tools
agent = Agent(
    llm = llm,
    description="You are my close friend",
    skills=[
        "You can remember all memory related to us",
        "You can remind the memory to answer questions",
        "You can remember the history of our relationship"
    ],
    memory_path=Path('templates/memory.json'),
    is_reset_memory=True # Will reset memory each time re-initialize agent. Default is False
)

# Step 2: invoke the agent
message = agent.invoke("Hello how are you?")
message.content

6. License

agentcore is released under the MIT License. You are free to use, modify, and distribute the code for both commercial and non-commercial purposes.

7. Contributing

We welcome contributions from the community. If you would like to contribute, please read our Contributing Guide. If you have any questions or need help, feel free to join Discord Channel.