Published
Graph Router
Graph-based routing — models query-LLM relationships as a graph for sophisticated routing decisions.
Overview
The Graph Router models the relationships between queries and LLM capabilities as a graph structure, enabling sophisticated routing decisions that capture complex interdependencies.
How It Works
Queries and LLMs are represented as nodes in a bipartite graph. Edge weights capture historical performance, query similarity, and model capability overlaps. Graph algorithms and GNNs are used to propagate information and make routing decisions.
Strategy
Models query-LLM relationships as a graph for routing decisions.
API Endpoint
autoroute:graphrouter
Use Cases
- Complex relationships between queries and LLM capabilities
- When model interactions and overlaps matter
- Multi-hop reasoning for routing
Best Practices
Graph Construction
The quality of the graph structure determines routing performance. Include both query-query similarity edges and query-model performance edges for the best results.
Related Models
- GMT Router — For graph-based personalized routing
- Causal LLM Router — For language understanding-based routing