Cut your n8n AI workflow costs by up to 60% with intelligent model routing

Hello guys !

I’m new to the n8n community but I’ve been building something that I think could be really useful for anyone running AI-heavy workflows here.

I built https://usekestrel.io - an open-source LLM routing proxy that sits between your app and your AI providers. It classifies each request in under 2ms and automatically routes it to the cheapest model that can handle it well. Instead of defaulting to GPT-4o or Claude for everything, Kestrel figures out when a cheaper model will do the job just as well.

I came across the n8n community while looking for places where people are building real AI workflows, and it seemed like cost management across multiple LLM providers is a common challenge here - so I wanted to share this and get your thoughts.

How it works with n8n

The integration is simple, just change the base URL in your OpenAI credentials or HTTP Request node:

Option 1: OpenAI credential settings

In your n8n OpenAI credential, set the base URL to:

https://api.usekestrel.io/v1

Option 2: HTTP Request node

If you’re calling /v1/chat/completions directly, just swap the URL:

https://api.usekestrel.io/v1/chat/completions

That’s it. Your existing workflows, prompts, and logic stay exactly the same. Kestrel handles the routing behind the scenes.

What it does

- Intelligent routing: ML classifier decides per-request whether to use GPT-4o, Claude, Gemini, Groq, Mistral, or a cheaper alternative based on complexity

- Semantic caching: identical or near-identical requests get cached responses, saving both time and money

- Real-time dashboard: see exactly where your money is going, per model and per workflow

- You keep your own API keys: Kestrel never stores or accesses your provider credentials directly

Pricing

This was important to me: you only pay 15% of what Kestrel saves you. If it doesn’t save you

anything, you pay $0. No monthly fees, no lock-in.

Links

- Website: https://usekestrel.io

- GitHub (open-source core): GitHub - andber6/kestrel · GitHub

Would love your feedback

Since I’m new here, I’d genuinely appreciate hearing from people who actually build these

workflows daily:

- How are you currently managing costs across multiple AI models?

- Would automatic routing be useful, or do you prefer choosing models manually per node?

- Any concerns or feature requests?

Happy to answer any questions!