domain-router

Domain Router

Intelligent query classification that routes user queries to specialized agents.

What It Does

The Domain Router is a Tool that:

1. Analyzes user query intent
2. Classifies into a domain (agent name or "not-related")
3. Stores classification for downstream agents
4. Runs once per query to prevent infinite loops

User Query → Domain Router → Classification → Agent Selection

Use When

The Domain Router is automatically invoked when:

• Custom agents are available for the user
• Query may require specialized knowledge
• Decision tree evaluates available tools

How It Works

Classification Flow

1. Router receives user query
2. Loads custom agent names and descriptions from hidden_environment
3. Creates dynamic prompt with available domains
4. LLM classifies query into one domain
5. Stores classification in hidden_environment
6. Subsequent agents check classification

One-Time Execution

The router only runs once per query:

async def is_tool_available(self, tree_data, ...):
    # Check if classification already exists
    classification_exists = (
        "domain_classification" in tree_data.environment.hidden_environment
    )
    # Only available if no classification yet
    return not classification_exists

This prevents infinite loops where the router would run on every decision tree iteration.

Classification Output

The router produces a DomainClassification result:

DomainClassification(
    domain="Immigration Law Expert",  # or "not-related"
    reasoning="Query asks about visa requirements which matches Immigration Law Expert",
    confidence="high"  # or "medium", "low"
)

Storage Location

Classification is stored in hidden_environment:

tree_data.environment.hidden_environment["domain_classification"] = {
    "domain": "Immigration Law Expert",
    "reasoning": "Query asks about visa requirements...",
    "confidence": "high"
}

Valid Domains