Hey Product Hunt community! 👋 I spent a lot of time thinking about the mounting expenses with AI over the last few months and came to the conclusion that routing specific tasks to the most optimal model is the absolute best way forward. You shouldn't have to keep an expensive cloud GPU idling 24/7 or write endless custom wrappers just to connect an LLM to your database or tools. That’s exactly why I built Mycelis. It’s an infrastructure-as-a-service platform that provisions open-source models on-demand, manages sandbox environments, and handles per-minute usage billing completely under the hood. The tech stack relies on a .NET 10 backend, PostgreSQL, Qdrant for semantic caching, and a Node.js gateway for Model Context Protocol (MCP) tool integrations. It works entirely as a secure proxy API gateway, meaning you can plug it straight into coding clients like OpenCode or any desktop tool. Everything is fully live in production. What do you think about the hybrid model approach (local IDE + serverless cloud tools)? Would love to get your brutal feedback, feature requests, or bug reports directly at https://mycelis.ai/feedback!

Mycelis

Serverless AI workspace with smart routing & MCP agents

Serverless AI workspace with smart routing & MCP agents