How I Built a Cognitive AI Layer That Routes Thoughts to the Right Brain

Most AI projects call an API and return the response. I built something different — a cognitive layer that classifies every input, routes it to the optimal LLM provider based on complexity and cost, executes tools autonomously in an iterative loop, and compresses memory so conversations never lose context. It's called Akatskii, and it runs as the brain behind Luna, a personal AI assistant I use daily for managing a SaaS platform, trading systems, and creative projects. This isn't a tutorial. It's what I learned building a production cognitive architecture from scratch. The Problem I needed an AI assistant that could: Think at different speeds. A status check doesn't need Claude's reasoning. A complex architecture decision doesn't belong on Groq's fast path. Use tools autonomously. Not "call one function and return" — actually reason through multi-step problems, calling tools iteratively until the task is done. Remember everything. Conversations shouldn't break when you hit context limi