jan as a router.

[tostane]

I was thinking about this idea claude as I have a small token account and wondered if this was possible?

I was hoping i could get some llms set to swap in as needed i was thinking to use claude to determine which was needed

11:19 PM
That's a smart architecture — using Claude as a "router" to dispatch tasks to the right model based on complexity, cost, or capability. Jan.ai supports multiple models so it's a good fit for this.

The basic idea would look like:

Router Claude (fast/cheap — Haiku) receives the request, classifies it, then hands off to the right model:

Simple Q&A / quick tasks → Haiku (fast, cheap)
Reasoning / coding / analysis → Sonnet
Complex multi-step / hardest problems → Opus
Local/offline tasks or privacy-sensitive → a local model like Llama or Mistral running in Jan
What the router prompt might look like: You'd give Claude a short system prompt like:

"Classify this user request as: simple, moderate, or complex. Respond with only one word."

Then your app logic reads that and routes accordingly.

The catch with Jan.ai is that it's primarily a chat UI — the model-swapping logic would need to live in something you build on top of the API directly (a small script, or a tool like n8n, LangChain, or a simple Python script) rather than inside Jan itself.

Are you thinking of building this as a custom script, or were you hoping Jan could handle the routing natively? That'll shape what the best approach looks like.