Model routing

Model routing

AI

Model routing sends each step of an AI agent's work to the best-suited model, such as a fast, low-cost one for simple steps and a stronger one for hard ones.

What is Model routing?

Not every step of a job needs the most capable model. Classifying a message or pulling a field from an order is simple; planning a multi-step job or writing a careful reply to an upset customer is not. Model routing matches the step to the model. Routing can be fixed or dynamic. Fixed routing assigns a model per agent or per stage of the work. Dynamic routing chooses at run time based on the request. Either way, the goals are lower cost, lower latency and keeping quality where it matters. In Concierge, each agent has a default model, either Chargezen Managed or one on your own provider account, and model overrides can use a different model for specific parts of the agent's work.

Example

For example, a support agent might use a fast model to read and sort incoming messages and a stronger model to plan a return that involves several items and a partial exchange.

Why Model routing Matters

Routing keeps quick answers fast and cheap while saving the stronger, costlier model for the work that needs it.

Model routing FAQ

Why not use the best model for everything?

Stronger models are usually slower and cost more per step. Most steps in a job are simple, so routing them to a faster model saves time and money without hurting the result.

Can I choose which model my agent uses?

In many agent products you can. In Concierge you pick a default model for each agent and can override it for specific parts of its work.

Explore More Terms

Browse our complete glossary of e-commerce and subscription terms.

View Full Glossary
Space, at your fingertips

What would you do with 22% more revenue?