Use Cases

from marsys.coordination import Orchestra
from marsys.coordination.topology.patterns import PatternConfig

# Research team with specialized agents
topology = PatternConfig.hub_and_spoke(
    hub="ResearchCoordinator",
    spokes=["WebSearcher", "PaperAnalyzer", "FactChecker", "ReportWriter"],
    parallel_spokes=True
)

result = await Orchestra.run(
    task="Research latest advances in quantum computing",
    topology=topology
)

topology = {
    "agents": ["User", "L1Support", "L2Support", "TicketManager"],
    "flows": [
        "User <-> L1Support",
        "L1Support -> L2Support",  # Escalation
        "L2Support -> TicketManager",
        "TicketManager -> User"
    ]
}

result = await Orchestra.run(
    task="Customer issue: Cannot login to account",
    topology=topology
)

# Specialized code reviewers
topology = PatternConfig.pipeline(
    stages=[
        {"name": "syntax", "agents": ["SyntaxChecker"]},
        {"name": "security", "agents": ["SecurityAuditor"]},
        {"name": "performance", "agents": ["PerformanceAnalyzer"]},
        {"name": "style", "agents": ["StyleReviewer"]},
        {"name": "summary", "agents": ["ReviewSummarizer"]}
    ],
    parallel_within_stage=False
)

result = await Orchestra.run(
    task=f"Review this code:\n{code_content}",
    topology=topology
)

from marsys.agents import Agent, AgentPool
from marsys.models import ModelConfig

# Create model config
config = ModelConfig(
    type="api",
    provider="openrouter",
    name="anthropic/claude-opus-4.6",
    temperature=0.3
)

# Create pool for parallel analysis
analyst_pool = await AgentPool.create_async(
    agent_class=Agent,
    num_instances=5,
    model_config=config,
    goal="Analyze financial markets and trends",
    instruction="You are a financial analyst. Analyze market data and provide insights.",
    name="FinancialAnalyst"
)

topology = {
    "agents": ["MarketMonitor", "AnalystPool", "RiskAssessor", "ReportGenerator"],
    "flows": [
        "MarketMonitor -> AnalystPool",
        "AnalystPool -> RiskAssessor",
        "RiskAssessor -> ReportGenerator"
    ]
}

topology = PatternConfig.pipeline(
    stages=[
        {"name": "research", "agents": ["TopicResearcher"]},
        {"name": "outline", "agents": ["OutlineCreator"]},
        {"name": "writing", "agents": ["ContentWriter", "TechnicalWriter"]},
        {"name": "editing", "agents": ["Editor", "FactChecker"]},
        {"name": "seo", "agents": ["SEOOptimizer"]},
        {"name": "publishing", "agents": ["Publisher"]}
    ],
    parallel_within_stage=True
)

# Hub-and-spoke for coordinated research
topology = PatternConfig.hub_and_spoke(
    hub="Coordinator",
    spokes=["Researcher1", "Researcher2", "Synthesizer"],
    parallel_spokes=True
)

# Pipeline for sequential validation
topology = PatternConfig.pipeline(
    stages=[
        {"name": "input", "agents": ["Validator"]},
        {"name": "process", "agents": ["Processor"]},
        {"name": "verify", "agents": ["Verifier"]},
        {"name": "output", "agents": ["Formatter"]}
    ]
)

# Mesh for collaborative decisions
topology = PatternConfig.mesh(
    agents=["Analyst1", "Analyst2", "Analyst3", "DecisionMaker"],
    fully_connected=True
)

# Hierarchical for tiered support
topology = PatternConfig.hierarchical(
    tree={
        "Manager": ["Supervisor1", "Supervisor2"],
        "Supervisor1": ["Agent1", "Agent2"],
        "Supervisor2": ["Agent3", "Agent4"]
    }
)