Autogen Conversation Agent Builder

# autogen_actors.py - AutoGen v0.4 Actor Model
from autogen_agentchat.agents import AssistantAgent, UserProxyAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_ext.models import OpenAIChatCompletionClient
from autogen_core.application import SingleThreadedAgentRuntime
from autogen_core.base import MessageContext
import asyncio

class ConversationOrchestrator:
    def __init__(self):
        self.runtime = SingleThreadedAgentRuntime()
        self.model_client = OpenAIChatCompletionClient(
            model="gpt-4",
            api_key="your-api-key"
        )
    
    async def create_research_team(self):
        """Create a team of specialized agents"""
        
        # Research Agent - Information gathering
        researcher = AssistantAgent(
            name="Researcher",
            model_client=self.model_client,
            system_message="""You are a research specialist who gathers 
            comprehensive information on technical topics. You provide detailed, 
            accurate information with citations.""",
            tools=[
                self._create_web_search_tool(),
                self._create_documentation_tool()
            ]
        )
        
        # Analyst Agent - Critical analysis
        analyst = AssistantAgent(
            name="Analyst",
            model_client=self.model_client,
            system_message="""You are a critical analyst who evaluates 
            information for accuracy, completeness, and practical applicability. 
            You identify gaps and inconsistencies."""
        )
        
        # Synthesizer Agent - Creates actionable output
        synthesizer = AssistantAgent(
            name="Synthesizer",
            model_client=self.model_client,
            system_message="""You are a synthesis expert who combines 
            research and analysis into clear, actionable recommendations. 
            You create structured, practical outputs."""
        )
        
        # User Proxy - Represents the user
        user_proxy = UserProxyAgent(
            name="User",
            code_execution_config=False
        )
        
        # Create group chat with round-robin pattern
        team = RoundRobinGroupChat(
            participants=[researcher, analyst, synthesizer, user_proxy]
        )
        
        return team
    
    def _create_web_search_tool(self):
        """Create web search tool for research agent"""
        async def web_search(query: str) -> str:
            """Search the web for information"""
            # Implementation using search API
            return f"Search results for: {query}"
        
        return web_search
    
    def _create_documentation_tool(self):
        """Create documentation lookup tool"""
        async def lookup_docs(topic: str, framework: str) -> str:
            """Look up official documentation"""
            # Implementation using docs API
            return f"Documentation for {topic} in {framework}"
        
        return lookup_docs
    
    async def run_conversation(self, task: str):
        """Execute conversational workflow"""
        team = await self.create_research_team()
        
        # Start conversation
        result = await team.run(
            task=task,
            max_turns=10
        )
        
        return result

# Usage
async def main():
    orchestrator = ConversationOrchestrator()
    
    task = """Research and analyze the best practices for implementing 
    microservices architecture with Node.js. Provide actionable 
    recommendations for a team of 10 developers."""
    
    result = await orchestrator.run_conversation(task)
    print(f"Result: {result}")

asyncio.run(main())

# python_agent.py - Python Agent in AutoGen v0.4
from autogen_core.application import SingleThreadedAgentRuntime
from autogen_core.base import MessageContext, TopicId
from autogen_core.components import DefaultTopicId, TypeSubscription
from dataclasses import dataclass

@dataclass
class AnalysisRequest:
    """Message type for analysis requests"""
    code: str
    language: str
    analysis_type: str

@dataclass
class AnalysisResponse:
    """Message type for analysis responses"""
    issues: list
    recommendations: list
    score: float

class PythonAnalyzerAgent:
    """Python agent that analyzes code"""
    
    def __init__(self, runtime: SingleThreadedAgentRuntime):
        self.runtime = runtime
        
        # Subscribe to analysis requests
        self.runtime.subscribe(
            type_subscription=TypeSubscription(
                topic_type="analysis",
                agent_type="PythonAnalyzer"
            ),
            message_type=AnalysisRequest,
            handler=self.handle_analysis_request
        )
    
    async def handle_analysis_request(
        self, 
        message: AnalysisRequest, 
        ctx: MessageContext
    ) -> None:
        """Handle incoming analysis requests"""
        
        # Perform analysis
        issues = await self._analyze_code(
            message.code, 
            message.language
        )
        
        recommendations = await self._generate_recommendations(issues)
        score = self._calculate_quality_score(issues)
        
        # Send response
        response = AnalysisResponse(
            issues=issues,
            recommendations=recommendations,
            score=score
        )
        
        await self.runtime.publish_message(
            message=response,
            topic_id=TopicId("analysis_results", ctx.sender)
        )
    
    async def _analyze_code(self, code: str, language: str) -> list:
        """Analyze code for issues"""
        # Use AST parsing, linting tools, etc.
        return [
            {"type": "security", "severity": "high", "line": 42, 
             "message": "SQL injection vulnerability"},
            {"type": "performance", "severity": "medium", "line": 15,
             "message": "Inefficient loop detected"}
        ]
    
    async def _generate_recommendations(self, issues: list) -> list:
        """Generate fix recommendations"""
        recommendations = []
        for issue in issues:
            if issue["type"] == "security":
                recommendations.append({
                    "issue": issue["message"],
                    "fix": "Use parameterized queries",
                    "code_example": "db.execute('SELECT * FROM users WHERE id = ?', [user_id])"
                })
        return recommendations
    
    def _calculate_quality_score(self, issues: list) -> float:
        """Calculate overall quality score"""
        if not issues:
            return 10.0
        
        severity_weights = {"critical": 3, "high": 2, "medium": 1, "low": 0.5}
        penalty = sum(severity_weights.get(i["severity"], 1) for i in issues)
        
        return max(0.0, 10.0 - penalty)

// CSharpAgent.cs - .NET Agent in AutoGen v0.4
using AutoGen.Core;
using AutoGen.Messages;
using System.Threading.Tasks;

public record CodeReviewRequest(
    string Code,
    string Author,
    string PullRequestId
);

public record CodeReviewResponse(
    bool Approved,
    List<ReviewComment> Comments,
    string Reviewer
);

public class DotNetReviewerAgent : IAgent
{
    private readonly IAgentRuntime _runtime;
    
    public DotNetReviewerAgent(IAgentRuntime runtime)
    {
        _runtime = runtime;
        
        // Subscribe to review requests
        _runtime.Subscribe<CodeReviewRequest>(
            topic: "code_review",
            handler: HandleReviewRequest
        );
    }
    
    private async Task HandleReviewRequest(
        CodeReviewRequest message,
        MessageContext context)
    {
        // Perform code review
        var comments = await AnalyzeCode(message.Code);
        
        // Request analysis from Python agent (cross-language!)
        var analysisRequest = new AnalysisRequest(
            Code: message.Code,
            Language: "csharp",
            AnalysisType: "security"
        );
        
        await _runtime.PublishAsync(
            message: analysisRequest,
            topicId: new TopicId("analysis", "PythonAnalyzer")
        );
        
        // Wait for Python agent response
        var analysisResult = await _runtime.ReceiveAsync<AnalysisResponse>(
            topicId: new TopicId("analysis_results", this.Name),
            timeout: TimeSpan.FromSeconds(30)
        );
        
        // Combine local and Python analysis
        comments.AddRange(ConvertToComments(analysisResult.Issues));
        
        // Send review response
        var response = new CodeReviewResponse(
            Approved: analysisResult.Score >= 7.0 && comments.Count(c => c.Severity == "critical") == 0,
            Comments: comments,
            Reviewer: this.Name
        );
        
        await _runtime.PublishAsync(
            message: response,
            topicId: new TopicId("review_results", context.Sender)
        );
    }
    
    private async Task<List<ReviewComment>> AnalyzeCode(string code)
    {
        // .NET-specific code analysis
        var comments = new List<ReviewComment>();
        
        // Use Roslyn analyzers
        comments.Add(new ReviewComment
        {
            Line = 10,
            Severity = "medium",
            Message = "Consider using async/await pattern",
            Suggestion = "Make this method async for better scalability"
        });
        
        return comments;
    }
}

# autogen_studio_config.py
from autogen_studio import Studio, AgentConfig, WorkflowConfig

class AutoGenStudioWorkflow:
    def __init__(self):
        self.studio = Studio()
    
    def create_customer_support_workflow(self):
        """Create customer support workflow in AutoGen Studio"""
        
        # Define agent configurations
        triage_agent = AgentConfig(
            name="TriageAgent",
            type="assistant",
            llm_config={
                "model": "gpt-4",
                "temperature": 0.3
            },
            system_message="""You are a customer support triage specialist. 
            Categorize incoming requests as: technical, billing, or general inquiry."""
        )
        
        technical_agent = AgentConfig(
            name="TechnicalSupportAgent",
            type="assistant",
            llm_config={"model": "gpt-4", "temperature": 0.2},
            system_message="You are a technical support expert.",
            tools=["search_knowledge_base", "create_ticket", "escalate_to_engineer"]
        )
        
        billing_agent = AgentConfig(
            name="BillingAgent",
            type="assistant",
            llm_config={"model": "gpt-4", "temperature": 0.1},
            system_message="You are a billing specialist.",
            tools=["check_invoice", "process_refund", "update_subscription"]
        )
        
        # Define workflow
        workflow = WorkflowConfig(
            name="CustomerSupportWorkflow",
            description="Automated customer support with specialized agents",
            entry_point=triage_agent,
            routing_logic={
                "technical": technical_agent,
                "billing": billing_agent,
                "general": triage_agent
            },
            max_turns=15,
            human_in_loop=True,  # Require human approval for refunds
            termination_condition="user_satisfied or max_turns_reached"
        )
        
        # Deploy to Studio
        self.studio.deploy_workflow(workflow)
        
        return workflow
    
    def monitor_workflow_performance(self, workflow_id: str):
        """Monitor workflow metrics in real-time"""
        metrics = self.studio.get_metrics(workflow_id)
        
        return {
            'total_conversations': metrics.conversation_count,
            'average_resolution_time': metrics.avg_resolution_time,
            'satisfaction_score': metrics.csat_score,
            'escalation_rate': metrics.escalation_rate,
            'cost_per_conversation': metrics.avg_cost
        }

# group_chat_patterns.py
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import SelectorGroupChat
from autogen_agentchat.base import TerminationCondition
from autogen_ext.models import OpenAIChatCompletionClient

class CollaborativeAgentTeam:
    def __init__(self):
        self.model_client = OpenAIChatCompletionClient(
            model="gpt-4",
            api_key="your-key"
        )
    
    async def create_code_review_team(self):
        """Create collaborative code review team"""
        
        # Security Expert
        security_expert = AssistantAgent(
            name="SecurityExpert",
            model_client=self.model_client,
            system_message="""You are a security expert. Review code for 
            vulnerabilities: SQL injection, XSS, CSRF, insecure dependencies."""
        )
        
        # Performance Expert
        performance_expert = AssistantAgent(
            name="PerformanceExpert",
            model_client=self.model_client,
            system_message="""You are a performance optimization expert. 
            Identify bottlenecks, inefficient algorithms, memory leaks."""
        )
        
        # Architecture Expert
        architecture_expert = AssistantAgent(
            name="ArchitectureExpert",
            model_client=self.model_client,
            system_message="""You are a software architect. Review for 
            SOLID principles, design patterns, maintainability."""
        )
        
        # Create selector group chat (agents speak when relevant)
        team = SelectorGroupChat(
            participants=[
                security_expert,
                performance_expert,
                architecture_expert
            ],
            model_client=self.model_client,
            termination_condition=TerminationCondition.max_messages(20)
        )
        
        return team
    
    async def review_pull_request(self, pr_code: str):
        """Review PR using collaborative team"""
        team = await self.create_code_review_team()
        
        task = f"""
        Review this pull request code:
        
        {pr_code}
        
        Each expert should:
        1. Analyze from your domain perspective
        2. Identify specific issues with line numbers
        3. Provide actionable recommendations
        4. Rate severity (critical/high/medium/low)
        
        Collaborate to produce comprehensive review.
        """
        
        result = await team.run(task=task)
        
        return result

{
  "maxTokens": 4000,
  "temperature": 0.3,
  "systemPrompt": "You are an AutoGen v0.4 specialist focused on building conversation-based multi-agent systems with actor model architecture"
}