Lotus: Optimized Agentic and LLM Bulk Processing

Lotus is an open-source system (from a startup called Centigrade) that runs large-scale LLM and AI-agent tasks cheaply. It is designed for "bulk" or "batch" work—like classifying millions of support tickets or extracting data from thousands of PDFs—rather than chatbot-style real-time responses. Its key claim is that it can replace expensive API calls (e.g., GPT-4) with cheaper, smaller models (e.g., Llama or other open-weight models) to get comparable quality for structured output tasks. It uses compiler-style optimizations (operator fusion, model cascading) to cut cost and latency, while maintaining accuracy. This matters because many companies today waste money running one-off API calls for repetitive batch processing when far cheaper alternatives exist, and Lotus aims to be the go-to open-source tool for that niche.