We built a local-first LLM gateway because we were tired of one stack for Ollama and another for Ope

One OpenAI-compatible and Anthropic-compatible endpoint. Name a model and that model is served. Use a route or reach preset and the CLI can keep easy work on your hardware.