MatrixOneGit for Data
The Cloud-Native, Hyper-Converged Database
The industry's first database to bring Git-style version control to data. Unify transactional, analytical, vector, and full-text workloads in a single system — MySQL compatible, AI-native, cloud-native.
Git for Data is what lets Agents explore — safely
An autonomous agent is only as trustworthy as the data layer beneath it. Git for Data gives every agent an isolated branch to experiment in, and gives you snapshot, time-travel, and rollback so exploration never turns into risk.
Free to Explore
Every agent forks production data at zero cost and experiments inside a fully isolated sandbox — run wild, break nothing, and merge back only what works.
Safe by Design
Snapshot before every step, time-travel to any past state, roll back in one click, and keep an immutable audit trail — agents act on real data with zero risk.
Powering Every Data Application
Ingest multimodal data, AI-driven processing, run hybrid workloads
MatrixOne
One Database for Everything
Say goodbye to 4 databases, multiple ETL jobs, hours of data lag, and sync nightmares
MatrixOneManage Data Like Code
The first database with Git for Data — every data change is traceable, reversible, and collaborative
Instant Snapshots
Zero-copy snapshots in milliseconds, no storage explosion
Time Travel
Query data as it existed at any point in history
Branch & Merge
Test migrations and transformations in isolated branches
Instant Rollback
Restore to any previous state without full backups
Complete Audit Trail
Track every data change with immutable history
Built for the AI Era
Built-in vector search, full-text search, and Agent data sandbox — no external vector databases needed
Vector + Full-Text Hybrid Search
Built-in IVF/HNSW vector indexes and full-text search engine. Supports billions-scale vector retrieval, directly powering RAG applications and semantic search.
AI Agent Data Sandbox
Git for Data branching naturally provides isolated data environments for AI Agents — one branch per agent, no interference, safe merging.
Deep MOI AI Integration
As the core data engine of MatrixOne Intelligence, deeply integrated with MatrixPipeline, MatrixGenesis, and other AI components — providing a unified data foundation for enterprise AI applications.
from matrixone import Client
client = Client()
client.connect(database='demo')
# Vector search
query = [0.2, 0.3, 0.4, 0.25, 0.35]
results = client.query(
Article.title,
Article.embedding.l2_distance(query)
).filter(
Article.embedding.l2_distance(query) < 0.1
).execute()Key Features
Built for Consolidation, Scale, and Intelligence
Three-Layer Cloud-Native Architecture
Disaggregated storage-compute design — each layer scales independently
Compute Layer
- •CN Compute Nodes
- •Resource Group Isolation
- •Auto-Scaling
- •Hot Data Cache
Transaction Layer
- •TN Transaction Nodes
- •Raft Shared Log
- •Conflict Detection
- •HAKeeper
Storage Layer
- •S3 Object Storage
- •Unlimited Scaling
- •Hot/Cold Separation
- •Low Cost
Raft Shared Log
S3 Object Storage
Flexible Deployment Options
From single-node to distributed, from private cloud to public cloud
Standalone
Primary-replica architecture for small to medium workloads
Object Storage
Single node + S3 object storage, balancing simplicity and elasticity
K8s Distributed
Fully distributed deployment with unlimited horizontal scaling
Get Started with MatrixOne
Start your AI data journey today
Branching is not bolted onto the database. It grows inside it.
Git4Data treats a database as a repository and a table as a versioned object, exposing snapshot, branch, diff and merge through SQL extensions. Implemented in MatrixOne, it leans on immutable object storage and MVCC so the cost of these operations tracks the size of the change rather than the size of the data.
1,000 concurrent agents: MatrixOne in 400 seconds, DoltDB unfinished at two hours
Raise concurrency from five agents to a thousand and the gap stops being a multiple. MatrixOne finishes in 400 seconds; DoltDB does not finish inside two hours. Grow the data tenfold as well, to scale factor 1,000, and most workloads still scale sublinearly.
BranchBench: agentic branching workloads
Four end-to-end workflows — software development, failure reproduction, data cleaning and Monte Carlo tree search — with five concurrent agents running 20 steps each, over roughly 47 million rows at scale factor 100. Against DoltDB, warm:
| Workload | MatrixOne | DoltDB | Speedup |
|---|---|---|---|
| Software development | 122.1 s | 1,925.6 s | 15.8x |
| Failure reproduction | 198.9 s | 1,677.3 s | 8.4x |
| Data cleaning | 58.6 s | 1,084.2 s | 18.5x |
| Monte Carlo tree search | 39.8 s | 410.2 s | 10.3x |
Git4Data: Database-Native Version Control for AI Agents
Hongshen Gou, Zuyu Zhang, Yuze Sun, Peng Xu, Feng Tian, Long Wang, Jianguo Wang
Agents explore many candidate states of relational data in parallel, and each one has to stay isolated, reproducible and auditable — ideally through the same SQL used for everything else. Source-code version control does not scale to data this size; relational databases manage the data well but rarely expose branching, comparison and merging natively. Git4Data is that missing layer.
Read the paperProof from customers
Jiangxi Copper: In-Database AI Vision for Real-Time Furnace Intelligence
In flash smelting, concentrate melts in two to three seconds and quality depends on precise slag control. JXCC built its furnace-side AIoT platform on MOI — sensors and images in one engine, vision models running as in-database UDFs — cutting furnace-bottom copper loss by 80%.
StoneCastle: Aggregate Queries from 10-30 Minutes Down to Seconds
Fund flows between ~6,000 depositors and 1,000+ banks generate millions of new transaction records a month and hundreds of millions of rows in total. StoneCastle rebuilt the stack on MOI in two phases: an HTAP engine replacing MySQL, then natural-language querying on top.
ETAO Tech: Three Databases Collapsed Back Into One
To absorb high-frequency acquisition, deep traceability queries and semi-structured intermediate results at once, the MES had bolted InfluxDB and MongoDB onto MySQL — pushing work that belongs in a database up into the application. Migrating to MatrixOne returned it to a single database, in under a week.