User input
|
Intent Router
AI

Intent Router

The Problem

Every tool you give your LLM makes it slower and more expensive, as it processes every possible action for each user request – wasting tokens, time, and money. As you scale, the problem only worsens.

The Solution

Intent Router instantly understands what users want and provides only the relevant tools to your AI. It's like having a smart traffic controller managing your AI actions.

Matching actions to user input

Examples

32 available actions Total 1381 tokens

Each AI tool requires its own detailed instructions (schemas). Without intent routing, every request to an LLM processes all tool schemas—using 1,381 tokens each time.

Try a request to see how intent routing reduces overhead by only choosing the most relevant tools.

Develop Dynamically, Build for Production

In Intent Router's development mode, you can dynamically define tools (and schemas) just as you would when building an LLM with tool calling. But when you're ready to deploy, we make it possible to train a significantly smaller, production-ready router model that can run almost anywhere. Think of it as compiling your code for production.