I Found a Stateless LLM Runtime on GitHub — It Dynamically Loads Models Per Request

I was randomly browsing GitHub when I came across this project called Chameleon.
At first I thought it was just another LLM wrapper — but it turned out to be something very different.

🔗 https://github.com/megeezy/Chameleon

Thoughts

I haven’t run it yet, but the architecture alone is interesting.

Curious to see:

• how it performs under load
• whether cold starts become a bottleneck
• how far the routing system can go

If you’re into AI infra, this is definitely worth a look.