openGauss: Building an Open and Win-Win Open Source Database Root Community | GBase Technology Cloud Sharing Conference Expert Talk Series · Cai Yajie

At the 2026 GBase Technology Cloud Sharing Conference, Cai Yajie, Secretary-General of the openGauss community and General Manager of the openGauss division, delivered a keynote speech entitled “openGauss: Building an Open and Win-Win Open Source Database Root Community”. He systematically introduced the achievements of openGauss over six years of open source, elaborated on the community’s “1+2” development strategy, showcased the latest breakthroughs in kernel technology, AI-native capabilities, and multi-write databases, and shared practical experience in the ecosystem co-construction between GBase and openGauss.

Cai Yajie began his speech by introducing the current development status of the openGauss community. Since going open source in 2020, openGauss has grown into the most popular open-source database root community in China. As of March 2026, the community had over 5.95 million users, more than 8,000 continuous contributors, generating over 30,000 PRs and 26,000+ issues, with 901 organizational members.

In terms of ecosystem development, the community has formed a multi-tier product system comprising “community open-source editions, commercial distributions, and user self-built editions”. openGauss download volumes rank among the top in China’s open-source database space, and it has become the leading one of the three major open-source database paths. Developers are spread globally, and its technological influence continues to expand.

“1+2” Strategy: Building a New Future for openGauss

Cai Yajie elaborated on the forward-looking “1+2” development strategy for openGauss.

“1” Persistence
Persisting in a Long-term Technology Evolution Path

openGauss will continue to strengthen the kernel and build a high-performance, highly secure database foundation. Through DPA (hardware accelerator) offloading technology, complex AP aggregation SQL computations are offloaded from the CPU to specialized hardware, freeing up CPU resources. In a small-scale 8U16C scenario, CPU occupancy is reduced by 72.6%, memory by 53%, read IO by 45%, and write IO by 60%.

“2” Creations
Creating Two Major Strategic Directions

• Direction 1: Creating oGRAC Multi-write Database and Super Node

oGRAC (openGauss Real Application Cluster) is a shared-memory multi-primary database architecture. Through three-layer pooling (compute, memory, storage), it enables concurrent reads and writes across multiple nodes. Its core technologies include: a globally incrementing SCN for cluster transaction visibility judgment, in-place update-level MVCC for consistent reads of UNDO pages, and page transaction slot row locking to improve transaction parallelism efficiency. oGRAC enables fast node failover with RTO <10 seconds and RPO=0; all compute nodes share a single copy of storage, reducing storage resource consumption by over 50%; two-node TPCC performance reaches 4.7 million tpmC, outperforming the industry benchmark by more than 5%. The super node technology integrates multiple computing resources such as CPUs, NPUs, DPUs, and GPUs through high-speed interconnect buses, with an estimated two-node TPCC of 5.4 million tpmC, RTO <6 seconds, and a scaling factor greater than 0.85.

• Direction 2: Creating an AI-native Multimodal Database Foundation

It includes core capabilities such as the DataVec vector database, the oGMemory agent memory system, and the AI Pipeline.

AI-native Capabilities:
DataVec and oGMemory

DataVec Vector Database

openGauss deeply integrates the DataVec vector engine into the database kernel, supporting the storage and retrieval of vector data types. DataVec supports multiple vector index algorithms such as IVFFLAT, HNSW, IVFPQ, and HNSWPQ, as well as similarity computations like L2 distance, cosine distance, and inner product. Users can manipulate vector data using familiar SQL syntax, enabling fused queries of structured and unstructured data. In terms of performance, DataVec improves vector search performance by 20% and reduces memory consumption by 90% through the RabitQ+BM25 indexing technology. Combined with optimizations of Kunpeng NEON, SVE, and other instruction sets, vector search efficiency is further enhanced.

oGMemory—Long-term Memory System for AI Agents 

oGMemory is a native vector memory engine built on openGauss. It has been successfully adapted to mainstream AI Agent frameworks such as OpenClaw and Claude Code. It provides full lifecycle context management, covering context collection, filtering, writing, indexing, recall, compression, and archiving, forming a runtime closed loop. oGMemory offers 8 types of precise memory extraction capabilities, including user profile merging, preference aggregation, event appending, and skill and tool experience accumulation, among others. In actual tests, oGMemory improved long-sequence task accuracy by 37% and reduced input token consumption by 68%. Additionally, oGMemory supports multi-Agent collaboration, providing stable, efficient, low-latency memory storage and retrieval for AI agents.

AI Pipeline—One-stop AI Business Flow Without Data Leaving the Database

openGauss, through in-database integrated processing, simplifies complex AI applications that traditionally require 7 steps and 4 boundary systems into a 2-step, single-system SQL operation. Automatic data vectorization (ONNX), intelligent retrieval (vector + BM25), and LLM generation/RAG are all completed within the database, achieving zero data movement and real-time response, effectively avoiding the risk of sensitive data leakage and data consistency issues in traditional solutions.

GBase × openGauss: Joint Ecosystem Building Practice

Cai Yajie specifically shared the in-depth cooperation case between GBase and openGauss.

Since joining the openGauss community in 2021, GBase has continuously contributed to community technology. In 2023, GBase contributed the first version of the community vector database; from 2024 to 2025, it continuously enhanced DataKit migration capabilities and vector capabilities; in 2026, GBase 8c V6 passed the national security and reliability assessment, becoming one of the products in China’s distributed transactional database field to receive this authoritative certification.

GBase continues to engage in technology sharing and ecosystem building, including over 10 community events and 20+ technology dissemination activities. GBase 8c has achieved large-scale deployment in telecommunications operators such as China Mobile, China Telecom, and China Unicom, as well as in government, energy, finance, transportation, and other industries.

From six years of community accumulation since open source, to the clear path of the “1+2” strategy, from the technical breakthrough of the oGRAC multi-write database to the AI-native capabilities of the DataVec vector database and oGMemory agent memory system, openGauss is building a globally leading open-source database root community with an open attitude. Looking to the future, GBase will continue to work hand in hand with the openGauss community, deeply integrate into the community’s development philosophy of “co-building, sharing, and co-governance,” continuously drive independent innovation and large-scale implementation in China’s database industry, and contribute Chinese strength to the development of the global database industry.