Preface

Today, I am very happy to share with you the latest product roadmap of Milvus 2025. In the past year, RAG has become a new paradigm for unstructured data processing. To meet the dual needs of retrieval accuracy and efficiency in RAG scenarios, Milvus has launched innovative features such as model integration, full-text retrieval, and hybrid retrieval, which have won wide recognition from the community.

Today, with the explosion of AI model capabilities, the demand for multimodal and unstructured data processing is increasing, and application fields such as agents, autonomous driving, and embodied intelligence are booming. Milvus will continue to work with ecosystem partners to focus on the pain points of actual scenarios, promote the implementation of technology, and make progress together with the community and users.

For 2025, we have planned two major versions:

• Milvus 2.6 focuses on optimizing the system architecture: simplifying the deployment threshold, optimizing the write path, reducing storage costs, improving the ability to delete and modify large-scale data, while strengthening the efficiency and functionality of scalar data and full-text retrieval, and introducing more model support.

• Milvus 3.0 will focus on AI integration, advanced search, and data management functions, creating a vector data lake system for offline scenarios that supports massive data processing, and strive to make Milvus a leading unstructured data processing platform.

So, which features do you hope to be launched as soon as possible, and which products do you think are urgently needed? You are welcome to vote for them in the poll below. Each of your suggestions will have an important impact on the future direction of Milvus.

In addition to the content in the above vote, we sincerely hope to hear more real voices from users and developers:

• What technology not mentioned above do you hope to see in future versions?

• Which functions still have room for improvement?

• Which details are worth further exploration?