A Python library that learns, in production, which agent in a multi-agent system performs best for each request — backed by four contextual multi-armed bandit algorithms and three deployment modes (in-process, REST/MCP server, learning proxy). - View it on GitHub