April 2026 monthly summary for infiniflow/ragflow. Delivered GraphRAG Task Resume Checkpoint Loading, enabling loading of previously saved subgraphs directly from the document store and thereby reducing unnecessary LLM calls and boosting efficiency. Completed a refactor to remove Redis-based checkpoints in favor of direct docEngine queries, establishing a single source of truth for subgraphs. Fixed a source_id query format mismatch and aligned loading logic with the established backend pattern to improve reliability across Elasticsearch/Infinity/OceanBase backends. Enhanced RAPTOR integration with a per-doc has_raptor_chunks check to skip processing when chunks already exist. Added 10 unit tests for checkpoint resume scenarios; overall test suite now 617 tests pass. Overall impact: lower latency, reduced compute costs, and more reliable handling of long-running workflows with clearer separation of concerns between GraphRAG and RAPTOR.

In July 2025, the team shipped a major feature to scale document processing and enhance rendering quality for the RAG-Anything product, while strengthening reliability and developer experience. Key work centered on Batch Processing for Parallel Document Handling with Enhanced Markdown (MD) to PDF conversion, plus supporting improvements across logging, configurability, and documentation.