LOTUS: Optimized Agentic and LLM Bulk Processing

LOTUS makes agentic and LLM bulk processing fast, easy, and robust. It introduces semantic operators (e.g., map, reduce, filter primitives) for processing structured and unstructured data corpora at scale with parallel agents and LLM calls. LOTUS’ optimized query engine allows you to write declarative code for complex data processing tasks with higher accuracy and lower cost.

LOTUS supports two classes of semantic operators. Agentic semantic operators (corpus.agent(ops=[...])) run tool-using agents over a corpus and are built for complex or ambiguous tasks that benefit from multiple steps and tool calls. LLM semantic operators (sem_map, sem_filter, sem_agg, sem_join, …) invoke far fewer model calls per record and are ideal for well-defined tasks such as LLM-as-judge evaluation, document extraction, and unstructured data analysis.