router-builder

Semantic routing classifies queries into categories (command, agent, skill, workflow) using embeddings, then matches them to specific resources. The router-builder creates a two-tier hierarchy that uses embedding-based similarity for fast, consistent resource selection across your AI assistant.

Hierarchical routing first classifies queries into broad categories, then selects a specific resource within that category. This two-tier approach reduces ambiguity and improves accuracy compared to flat routing, especially when managing multiple resource types like commands, agents, and workflows.

Yes. The router-builder supports dynamic route addition via YAML configuration, allowing you to extend routing rules and add new resources without retraining the embedding model or modifying core routing logic.

The router-builder uses the same HuggingFaceEncoder model as your RAG pipeline, ensuring consistent embedding representations across both systems and improving routing accuracy and alignment with your retrieval logic.

Semantic routing excels when queries vary in phrasing but share intent, or when you need to scale routing across many resources. Rule-based routing works well for fixed, predictable patterns. Use semantic routing for flexible, intent-driven classification in multi-resource AI workflows.

Install semantic-router, sentence-transformers, and pyyaml via pip. Define your routes in YAML with categories and resources, then initialize the router with your HuggingFaceEncoder. Test with the provided test suite to verify routing behavior.