Enables Claude to route sub-tasks to the cheapest suitable model tier based on semantic rules, with tools for dispatching and previewing routing decisions.
Offline, zero-dependency tool-router for MCP that funnels N tools down to a ranked shortlist for a natural-language query, so the model reads ~8 tools instead of 200.
Enables agents to discover, evaluate, and serve task-specialist models (embeddings, reranking, classification, extraction) with OpenAI-compatible endpoints and MCP tools for routing, embedding, reranking, and extraction.
Enables step-level routing of AI workflows by decomposing tasks, selecting the best model per step within constraints, executing steps, and providing full execution traces.