What and how it can be used:
The LLM Router component intelligently selects and routes requests to the most appropriate language model based on query characteristics, complexity, cost considerations, or performance requirements. It dynamically chooses between different LLMs (GPT-4, Claude, Gemini, etc.) or model tiers to optimize for speed, cost, quality, or specialized capabilities.
When/how the component should be used:
- Use when multiple models are available and selection must be dynamic.
- Ideal for balancing cost and performance by routing simple queries to cheaper models.
- Essential for cost optimization in high-volume applications.
- Define a pool of available LLMs with known characteristics.
- Configure routing rules (e.g. task type, priority, cost tier).
- Route the prompt to exactly one selected model.
- Provide a fallback model in case the preferred model fails.
Connections with other components:
- Chat Output
- Text Input
- Text Output
- Agent Core
- API Request
- Directory
- News Search
- RSS Reader
- SQL Database
- Web Search
- Language Model
- LLM Router
- Batch Run
- Data Operations
- Parser
- Python Interpreter
- Save File
- Smart Function
- Split Text
- Structured Output
- Type Convert
- Listen
- Notify
- Smart Router
- Calculator
- Anonymization
- Guardrail
- Human-in-the-loop
- Bing Search API
- Google Search API
- ChromaDB
- If-Else
Configurable settings:
- Language Model ( from Language Model component)
- Input ( From another component or write the input)
- Judge LLM ( From Language Model component)
- Optimization (Choose option from list select)
Default settings:
- Language Model ( from Language Model component)
- Input ( From another component or write the input)
- Judge LLM ( From Language Model component)
- Optimization (Choose option from list select)
Control Section:
- Language Models
- Input
- Judge LLM
- Optimization
- Use OpenRouter Specs
- API Timeout
- Fallback to First Model
Default values:
- Optimization = Balanced
- Use OpenRouter Specs = on
- API Timeout = 10
- Fallback to First Model = on
Desired Behaviour:
- Deterministic routing
- Graceful fallback on failure