LightRAG's document processing pipeline now strips full document content from in-memory queue payloads, computing and persisting summary metadata upfront, and re-reading bodies from storage on demand to lower memory pressure.
Memory-optimized document pipeline that avoids holding full document bodies in in-memory processing queues by re-reading from persistent storage as needed.
Operators processing large document batches in LightRAG can expect lower peak memory usage during pipeline runs, reducing the chance of out-of-memory errors or needing larger instances. The optimization maintains the same functional output by deferring body reads to storage, but teams should monitor queue throughput and storage read latency under heavy load to confirm no new I/O bottlenecks appear.