router-builder

You can route natural language queries to specific agent workflows using a hierarchical semantic router that matches utterance embeddings against defined categories and domains to select the correct resource.

The best way to orchestrate task distribution among subagents is using a two-tier routing system that categorizes mixed utterances and directs them to the appropriate command, agent, or skill resource.

Embedding-based semantic routing uses a HuggingFace all-MiniLM-L6-v2 encoder to process slash commands and mixed utterances, matching their vector representations against YAML route definitions to determine the correct target.

Yes, you can dynamically add new routing paths for agent tasks via API, allowing you to inject new route definitions into the routing system on the fly without interrupting ongoing workflows.

Yes, you need YAML route definitions under the routing routes directory to configure the hierarchical router, as these files specify the categories and domains required for accurate semantic task distribution.

A two-tier category and domain routing system provides faster and more accurate resource selection than a flat router by narrowing down semantic matches hierarchically, enabling scalable decision-making across numerous agents.