3 lines. Any backend. Local or cloud. Build single agents or coordinated fleets with effort-based model routing, runtime backend switching, real memory, and zero lock-in. Run on your GPU, on Ollama, ...