Skip to main content

Intelligent Routing

NorthSignal's defining feature is its Intelligent Routing Engine. Instead of forcing you to choose the best AI model for every prompt, NorthSignal analyzes the complexity of your request and automatically sends it to the best expert for the job.


How the Router Decides

The routing engine evaluates your prompt across multiple dimensions in milliseconds:

Prompt CharacteristicRouted ToWhy
Coding, debugging, complex logicClaude Sonnet 4.5 (Anthropic)Best-in-class for software engineering tasks
Very long documents (100+ pages)Gemini 2.0 Pro (Google)Largest context window — can read entire books
Quick questions, creative writingGPT-5 (OpenAI)Fastest time-to-first-token, versatile
Simple tasks, summariesFastest available modelPrioritizes speed over depth

The router considers:

  • Token complexity — How sophisticated is the language and structure of the prompt?
  • Input length — How much text, code, or file data are you sending?
  • Task type signals — Keywords like "debug," "refactor," "summarize," or "compare" influence routing.
  • Available keys — The router only considers providers you've connected.
tip

The more API keys you add to NorthSignal, the smarter the routing becomes. With all three providers connected, the router has maximum flexibility to pick the optimal model.


Overriding the Router

You always have full control. If you want to use a specific model:

  1. Look below the chat input bar for the Model Selector dropdown.
  2. Click the dropdown — it defaults to Auto-Route.
  3. Select your preferred model from the list (e.g., Claude Sonnet 4.5).

NorthSignal will lock your current chat to that model until you change it back.


Why Not Just Always Use the "Best" Model?

Different models have different strengths and different pricing. The Intelligent Router saves you money by:

  • Sending simple questions to cheaper, faster models.
  • Only invoking expensive reasoning models (like o1) when the task truly demands it.
  • Picking the provider with the largest context window when you attach massive files, rather than hitting a context limit error.

The result is higher quality answers at lower cost — without you having to think about which model to pick.