Ai

RAGFlow v0.24.0 Adds Memory APIs, Persistent Agent Sessions, and New Data Connectors

RAGFlow v0.24.0 Adds Memory APIs, Persistent Agent Sessions, and New Data Connectors

RAGFlow v0.24.0 delivers a substantial platform update centered on developer extensibility, agent usability, and broader enterprise integration. The release introduces new Memory APIs and SDK support, a redesigned chat-style agent interface with persistent session history, and multi-sandbox support for more flexible deployment and execution. Alongside these additions, the update improves retrieval for deep research workflows and expands the ecosystem with new databases, OCR tooling, models, and data connectors.

What Changed in v0.24.0

A major addition in this release is the new Memory capability, which now includes APIs and an SDK for developer integration. RAGFlow also adds Memory extraction log visibility in the console, making debugging and traceability easier for teams building production-grade AI workflows.

On the dataset side, batch metadata management has been added, and the labeling of ToC has been renamed to PageIndex. For agent workflows, RAGFlow launches a new chat-like conversation management interface that preserves sessions and dialogue history, which should make ongoing agent interactions more practical for real-world use cases.

The release also introduces a multi-sandbox mechanism, with current support for local gVisor and Alibaba Cloud, while keeping compatibility with mainstream sandbox APIs through admin configuration. In chat, a new Thinking mode replaces the earlier Reasoning configuration, and retrieval strategies have been optimized for deep-research scenarios to improve recall accuracy.

Administrative and model management capabilities have also been expanded. Multiple Admin accounts are now supported, and the model configuration center includes a connection testing feature when adding new models. RAGFlow also adds support for OceanBase as an alternative to MySQL, PaddleOCR-VL for document and vision workflows, and new model options including Kimi 2.5, Stepfun 3, and doubao-embedding-vision.

From an integration perspective, the platform adds new data source connectors for Zendesk, Bitbucket, and others. The changelog also notes a fix for a knowledge graph search issue, signaling continued attention to search quality and platform stability.

Why It Matters

This release pushes RAGFlow further toward enterprise-ready AI infrastructure. The addition of Memory APIs and SDK support makes the platform easier to embed into custom developer workflows, while the improved logging helps teams monitor and troubleshoot memory-related behavior more effectively.

The new persistent agent session interface is particularly important for organizations building conversational assistants or internal AI tools that need continuity across user interactions. Combined with multi-sandbox support, the update also strengthens deployment flexibility for teams with stricter runtime isolation or cloud-specific requirements.

Support for OceanBase, PaddleOCR-VL, and new model providers broadens the platform’s compatibility across database, document processing, and model orchestration layers. Meanwhile, connectors like Zendesk and Bitbucket make the system more useful for support, engineering, and knowledge operations teams that want retrieval pipelines tied directly to their operational tools.

Overall, v0.24.0 is less about a single headline feature and more about platform maturity. It enhances memory tooling, agent continuity, retrieval quality, infrastructure choice, and enterprise connectivity in ways that should matter to teams deploying retrieval-augmented AI systems at scale.

Official Source: https://github.com/infiniflow/ragflow/releases/tag/v0.24.0