Top 10 Popular AI Agents & Frameworks

from langgraph.graph import StateGraph, END
from langgraph.checkpoint.memory import MemorySaver
from typing import TypedDict, Annotated
import operator

class AgentState(TypedDict):
    messages: Annotated[list[str], operator.add]
    next_step: str

def reason_node(state: AgentState):
    # LLM plans next action or terminates
    return {"messages": ["Agent reasoned next step."], "next_step": "act"}

def tool_node(state: AgentState):
    return {"messages": ["Tool executed successfully."], "next_step": "end"}

# Build Cyclical State Graph
workflow = StateGraph(AgentState)
workflow.add_node("reason", reason_node)
workflow.add_node("act", tool_node)
workflow.set_entry_point("reason")

workflow.add_conditional_edges(
    "reason",
    lambda state: "act" if state["next_step"] == "act" else END
)
workflow.add_edge("act", "reason") # Complete the reasoning cycle

app = workflow.compile(checkpointer=MemorySaver())

from crewai import Agent, Task, Crew, Process

# Define Specialized Agents
researcher = Agent(
    role="Senior Market Intelligence Researcher",
    goal="Extract quantitative metrics on frontier AI models",
    backstory="You are an expert tech analyst known for rigorous data verification.",
    verbose=True
)

writer = Agent(
    role="Technical Editorial Director",
    goal="Synthesize research into authoritative enterprise whitepapers",
    backstory="Former IEEE senior editor who translates complex architectures clearly.",
    verbose=True
)

# Assign Deliverable Tasks
task_research = Task(description="Analyze 2026 AI agent latency and token benchmarks.", agent=researcher)
task_write = Task(description="Draft comprehensive 1,500-word executive brief.", agent=writer)

# Assemble Hierarchical Crew
tech_crew = Crew(
    agents=[researcher, writer],
    tasks=[task_research, task_write],
    process=Process.hierarchical, # Manager LLM handles delegation
    manager_llm="claude-3-7-sonnet"
)
result = tech_crew.kickoff()

from autogen import AssistantAgent, UserProxyAgent

# Code-Generating Assistant
engineer = AssistantAgent(
    name="senior_engineer",
    llm_config={"model": "gpt-4o", "temperature": 0.0},
    system_message="Write bug-free Python scripts to solve data engineering tasks. Output ```python code blocks."
)

# Automated Code Executor (Docker or Local Sandbox)
executor = UserProxyAgent(
    name="sandbox_runner",
    human_input_mode="NEVER",
    code_execution_config={"work_dir": "workspace", "use_docker": True},
    max_consecutive_auto_reply=5
)

# Autonomous Execution & Self-Correction Turn
executor.initiate_chat(
    engineer,
    message="Fetch latest currency exchange rates via API, compute volatility index, and plot results to 'volatility.png'."
)

from agency_swarm import Agent, Agency, BaseTool
from pydantic import Field

# Type-safe custom tool
class DatabaseAuditTool(BaseTool):
    cluster_id: str = Field(..., description="PostgreSQL production cluster ID")
    
    def run(self):
        # Executes read-only schema analysis
        return f"Cluster {self.cluster_id}: 0 lock contentions detected."

# Define Agency Structure with restricted communications
ceo = Agent(name="CEO", description="Client facing manager receiving project briefs.")
analyst = Agent(name="DataAnalyst", tools=[DatabaseAuditTool], description="Database diagnostics specialist.")

# Explicit Communication Flow: User -> CEO -> DataAnalyst
agency = Agency(
    agency_chart=[
        ceo,            # Entry point
        [ceo, analyst]  # CEO can communicate with Analyst, but Analyst cannot talk directly to user
    ],
    shared_instructions="Always deliver audited production recommendations."
)

from camel.societies import RolePlaying
from camel.types import RoleType

# Initialize Inception Roleplay
task_prompt = "Architect an automated disaster recovery pipeline for multi-region Redis clusters."
role_play = RolePlaying(
    assistant_role_name="Cloud Infrastructure Architect",
    user_role_name="Chief Information Security Officer",
    task_prompt=task_prompt,
    with_task_specify=True
)

# Step Through Turn-Based Cooperative Dialogue
input_msg = role_play.init_chat()
for turn in range(4):
    assistant_response, user_response = role_play.step(input_msg)
    if "CAMEL_TASK_DONE" in assistant_response.msg.content:
        break
    input_msg = assistant_response.msg

from smolagents import CodeAgent, HfApiModel, tool

@tool
def query_vector_database(query: str) -> str:
    """Searches corporate knowledge base for architecture guidelines.
    Args:
        query: Search keywords
    """
    return "Kafka clusters require minimum 3 broker replicas for high availability."

# Instantiate Code-First Agent
agent = CodeAgent(
    tools=[query_vector_database],
    model=HfApiModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct"),
    max_steps=4
)

# Agent writes Python code to call query_vector_database and parse output
response = agent.run("Find our HA requirements for Kafka and compute recommended partition count for 10k eps.")

from multion import MultiOn

client = MultiOn(api_key="mo_live_...")

# Launch Autonomous Browser Session
session = client.sessions.create(
    url="https://cloud.provider.com/login",
    use_proxy=True
)

# Instruct Agent to Navigate and Extract Infrastructure Metrics
response = client.browse(
    session_id=session.session_id,
    cmd="Log into the analytics console, navigate to us-east-1 billing, and download the CSV invoice for September 2026."
)

print(f"Action Status: {response.status}")
print(f"Extracted Artifacts: {response.data}")

import anthropic

client = anthropic.Anthropic()

# Configure Computer Use Tool with Display Dimensions
response = client.beta.messages.create(
    model="claude-3-7-sonnet-20250219",
    max_tokens=2048,
    tools=[
        {
            "type": "computer_20241022",
            "name": "computer",
            "display_width_px": 1024,
            "display_height_px": 768,
            "display_number": 1
        }
    ],
    messages=[
        {
            "role": "user",
            "content": "Open LibreOffice Calc, create a financial balance sheet for Q3, and save to '/reports/q3_balance.xlsx'."
        }
    ]
)
# Claude responds with actions: {"action": "mouse_move", "coordinate": [450, 220]} or {"action": "type", "text": "Q3 Revenue"}