Technical deep dives, product updates, and practical engineering insights for Data + AI teams
Part 2 of the MatrixOne Git4Data series: a hands-on, copy-paste-runnable walkthrough. Install MatrixOne, load a million rows, then run every Git primitive — snapshot, clone, branch, row-level diff, merge with conflict modes, cherry-pick, and PITR — across table, database, account, and cluster levels, with measured numbers showing version-control cost is independent of data size.


A Snowflake veteran's notes from Summit 2026: how the company finally made its AI label stick — and why, after a long detour, its real moat as a data company is shining for the first time in the AI era.

Part 1 of the MatrixOne Git4Data series: why version control was the underrated engine of software productivity, why data is still stuck in the 'SVN era,' and how MatrixOne makes branch, merge, and restore cheap on TB-scale data — the Git moment for data at scale.

MatrixOne integrates NVIDIA cuVS and RAFT directly into the database engine, enabling GPU-accelerated vector indexing and retrieval at enterprise scale. This article explains the architecture, implementation details, and benchmark results across datasets up to 88 million vectors, demonstrating significant gains in indexing speed, query throughput, and memory efficiency.

We explore MatrixOrigin's journey toward becoming an AI-Native organization, and how GitHub, AI Agents, and end-to-end ownership are reshaping collaboration, productivity, and organizational design.

How MOI's intelligent CV-screening solution connects the entire first-pass screening pipeline — multimodal parsing, JD alignment, intelligent scoring and ranking, and dashboard output — making screening efficient, standardized, and reusable, helping enterprises raise both the speed and quality of first-pass screening in high-volume hiring.

Using the store-app project as a case study, exploring the real-world experience, value, and limitations of Memoria in software development.

This article shows how the MOI sourcing and price comparison Agent connects procurement analysis end to end, from requirement structuring and multi-source data fusion to supplier evaluation and automated report generation.

Against the backdrop of the industry's shift from GenAI to AI Agents, this article reviews the team's five-stage journey from "integrating products into the AI ecosystem" to "adapting the organization to the AI era," analyzes the tension between individual productivity explosions and organizational collaboration bottlenecks, and distills core insights such as "AI is a magnifying glass, not a money printer.

Current independent AI Agent deployments can easily create new intelligent silos. This article analyzes core challenges from three perspectives: data sharing, data generated by Agents, and multi-type data fusion. A unified data platform is critical to avoiding duplicated construction, enabling experience reuse, and enforcing permission control.

This article explains three major pain points in AI Agent development: process crashes, context window overflow, and memory loss after model switching. It proposes a death note mechanism and combines it with Memoria to preserve instant Agent state and reuse long-term memory, supporting crash recovery, context compression, and seamless model switching for more stable long-running tasks and multi-model collaboration.

Databases have not disappeared. Instead, they are undergoing a paradigm shift in the AI era. This article reviews the evolution of data abstraction layers, analyzes Markdown's role in Agent memory, and discusses the trend toward unified cognitive-state infrastructure for multi-Agent systems.

Memoria officially launches Agent memory backup and restore, supporting point-in-time snapshots of Agent memory and one-click instant recovery when issues occur. The free version provides two snapshot slots.