BETA · privacy LLMs & voice servers operational · GPU upgrade underway for faster responses · packages may still change Status & Roadmap →

Intelligent AI routing: how the system automatically chooses the right model

Hybrid Routing analyses every question and automatically routes it to the most suitable AI model. Simple questions go to a lightweight local model; complex questions to a premium model. Save costs without sacrificing quality.

← ← Back to knowledge base

What is Hybrid Routing?

Hybrid Routing is an intelligent model-selection layer that automatically determines the most suitable AI model for every incoming question — maximising quality for complex queries while minimising costs for simple ones.

How it works

For each question, the router analyses: question complexity, sensitivity (medical, legal, financial terms), and model availability. Based on these signals it routes to the appropriate model from your configured model pool.

Configuration

  1. Set the budget model pool for routine questions (e.g. Mistral 7B, GPT-4o Mini).
  2. Set the premium model pool for complex questions (e.g. Claude 3.5 Sonnet, GPT-4o).
  3. Define the complexity threshold — from "aggressive saving" to "quality first".

Savings potential

On average 60–70% of customer questions are routine and can be excellently handled by a lightweight model. Hybrid Routing can reduce credit costs for those questions by 50–80% compared to always using a premium model.