Can AI chats automatically choose the right model for each task—and show whether it actually saves money?

Hi everyone, I’m Miha Tavčar, founder of Korako Yolawani, a personal AI assistant currently in development.

A user may stay on a powerful, expensive model throughout a conversation, even when a simpler model could handle some requests reliably.

Does anyone know of—or use—a chat interface that automatically routes each request to an appropriate model within the same conversation, preserving relevant context without requiring manual switching?

I’m especially interested in three things:

1. Reliability: How does it decide that a cheaper model is sufficient, and detect when a stronger model is needed?

1. Real savings: Does it measure the total cost of a successfully completed task, including routing, context transfer, retries, and verification?

1. User control: Can users see which model was used and why, set spending limits, and require approval before a more expensive escalation?

I’m exploring this for Korako Yolawani. The aim is to make cost-conscious AI use easier for people who shouldn’t need to understand every model or monitor every token.

What have you tried, what savings have you measured, and where does the approach still fail?

If you know an existing tool, open-source project, or research paper—or are building something similar—I’d love to hear about it. Please share a link or example in the comments, or feel free to message me directly here on AI Lounge. I’d be happy to connect and learn from your experience.

Thank you!
Miha Tavčar
Founder, Korako Yolawani