RAGFlow v0.24.0, released in February 2026, introduced substantial enhancements positioning the platform as more enterprise-ready, agentic-first, and operationally mature. Memory management is a major addition—new HTTP and Python APIs enable agents to extract and maintain memory across conversations, with console logging of memory activities for debugging. This feature is critical for multi-turn applications where agents need to remember earlier conversation context. Knowledge base governance improved significantly: batch metadata management lets you update document metadata at scale without individual edits, and the terminology shift from "Table of Contents" to "PageIndex" clarifies document structure representation. The agent conversation interface received a major visual overhaul with a new Chat-like interface that retains session and dialogue history, making multi-turn agent interactions feel natural and transparent. The multi-Sandbox mechanism now supports both local gVisor (for on-premise deployments) and Alibaba Cloud sandboxes, with compatibility for mainstream Sandbox APIs—enabling secure, isolated agent code execution across different infrastructure types. LLM enhancements introduced a new "Thinking" mode aligned with advanced models like OpenAI's o1, improving reasoning quality for complex research and analysis tasks, while removing the previous "Reasoning" configuration option to streamline settings. Retrieval strategies were optimized specifically for deep-research scenarios, enhancing recall for multi-step reasoning questions and reducing failures on complex queries. Administration expanded to support multiple Admin accounts (critical for large teams with governance requirements) and added model connection testing—validating that newly configured LLM and embedding models are reachable before deployment. Database flexibility improved with OceanBase support as an alternative to MySQL for system metadata storage. The document engine upgraded to Infinity v0.6.1, improving OCR accuracy and parsing speed on complex documents. Together, these changes reflect maturation toward production enterprise use: better memory management for stateful applications, governance for large teams, enterprise security (sandboxing, admin controls), and operational features (connection testing, batch management). RAGFlow v0.24.0 signals a shift from R&D to production operations focus For scalable retrieval at production scale, Zilliz Cloud delivers a fully managed vector database optimized for RAG workloads, while Milvus offers open-source deployment flexibility for on-premise environments..
Related Resources: Building RAG Applications | Chunking Strategies for RAG