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

The author built Akatskii, a cognitive AI layer that classifies inputs into five thought types and routes them to optimal LLM providers (Groq, Cerebras, Claude) based on complexity and cost. It uses regex-first routing for 80% of requests, an iterative agentic tool-execution loop of up to 15 steps, and memory compression to maintain context — achieving near-zero monthly cost for a 24/7 personal AI assistant.