OpenLumara: token-efficient AI agent for local models

View on Reddit

The agent uses an extremely small system prompt and is designed for token efficiency. It is written from scratch, not 'vibecoded', and runs on modest hardware. The framework is modular and made for local models.