Triage Router
Not every question needs the expensive AI. See how a router sends easy questions down a fast lane, and saves the slow, costly model for the hard ones.
New here? Read the short guide for this demo, with steps to follow and what to look for.
The idea in one minute
Think of a help desk.
At a good help desk, a front-desk person answers most questions quickly. Only the tricky ones go to a senior specialist, who is slower and costs more. Nobody sends "what are your opening hours?" to the specialist.
AI works the same way. A router reads each question first and picks a lane: a fast, cheap model for easy questions, or a slower reasoning model for hard ones. Try it below.
- 1Type a question, or pick an example.
- 2See which lane it takes, and why.
- 3Move the dial to see the savings and the risks.
This is a demo with made-up data, and it runs only in your browser. The questions are invented samples. No AI model is called, nothing you type is sent anywhere, and nothing is stored. The costs and times are illustrative numbers for a "fast" and a "reasoning" model, not measurements and not any company's prices.
1 Try a question
The coloured bar is this question's difficulty. The dark tick marks where the thinking lane starts.
What the router noticed
2 How careful should it be?
Lower the dial and more questions get the careful, expensive treatment. Raise it and you save more, but hard questions may slip into the fast lane.
3 What it does to 20 sample questions
Click any question below to try it in step 1.
How the router decides
Clues add up to a score.
- Long or multi-part questions earn a point or two.
- Words like "explain", "compare", "why" or "plan" ask for reasoning, so they earn more.
- Errors, code and high-stakes topics such as legal or tax raise the score.
- Greetings and stock questions lower it. At the dial's number of points or more, the question goes to the thinking lane.
Honest limits
What this demo does not do.
- The clues are simple word rules. A real router often uses a small trained classifier, or lets the fast model say how confident it is.
- Real systems also escalate: if the fast model is unsure or its answer fails a check, the question is retried on the reasoning model.
- The cost and time figures are illustrative. Real prices and speeds depend on the models and your traffic.
Have an idea? Let's build it properly.
Tell us what you want to build or automate. We'll reply with practical next steps.