# MatrixOrigin

> Enterprise infrastructure for AI agents — **Control Your Data, Trust Your AI**. MatrixOrigin builds an AI-native data stack that turns enterprise data into AI-ready assets and runs trustworthy AI agents in production.

MatrixOrigin is an enterprise AI infrastructure company. Its stack spans four products — a cloud-native database, enterprise AI infrastructure for agents, an agent runtime platform, and agent memory — so enterprises can go from raw multi-source data all the way to production AI agents on one trusted foundation.

Primary sites: English [www.matrixorigin.io](https://www.matrixorigin.io) · Chinese [matrixorigin.cn](https://matrixorigin.cn). Documentation: [docs.matrixorigin.cn/en](https://docs.matrixorigin.cn/en/).

## Products

- [MatrixOne](https://www.matrixorigin.io/matrixone): **Git for Data.** The cloud-native, hyper-converged database — snapshot, branch, merge, time-travel and rollback your data like code. Unifies transactional, analytical, vector and full-text workloads in one MySQL-compatible, AI-native system. Branching gives every AI agent an isolated sandbox to experiment in, while snapshots, time travel and an immutable audit trail keep that exploration risk-free.
- [MatrixOne Intelligence (MOI)](https://www.matrixorigin.io/moi): **Enterprise AI infrastructure for agents.** Unifies enterprise data, model services, business context, agent execution and long-term memory. Multimodal parsing, knowledge retrieval and data governance make enterprise information usable as agent context. Workflow orchestration and governed tool use connect AI capabilities to business processes, with access isolation and execution traceability for building, running and managing production agents.
- [Astra](https://www.matrixorigin.io/astra): **The agent runtime platform for complex work.** Agents that don't just answer once, but keep working toward a goal — understand the objective, build a trackable task graph, call governed capabilities, keep the evidence, resume after waiting/failure/disconnection, and explain the whole thing afterwards. Built on one shared runtime backbone with many capability providers.
- [Memoria](https://thememoria.ai): **Persistent memory for AI agents.** Long-term, adaptive, enterprise-grade memory so agents remember, learn and improve across sessions.

## MatrixOne Intelligence capabilities

- [Data Lakehouse](https://www.matrixorigin.io/moi/data-lakehouse): Cloud-native lakehouse for all workloads.
- [Data Governance](https://www.matrixorigin.io/moi/data-governance): Unified governance, lineage tracking and compliance.
- [Data Pipeline](https://www.matrixorigin.io/moi/data-pipeline): Multi-source connectors with visual ETL orchestration.
- [Document Intelligence](https://www.matrixorigin.io/moi/document-intelligence): Smart document parsing, extraction and knowledge base construction.
- [Agentic Data Search](https://www.matrixorigin.io/moi/agentic-data-search): Natural language retrieval and analytics across all data sources.
- [AI Agent Builder](https://www.matrixorigin.io/moi/ai-agent-builder): Build production AI agents with tool use.

## How Astra works

- **One runtime backbone, many capability providers.** Web, CLI, SDK, enterprise apps and edge executors don't each implement their own agent — they share the same session, run, Work, context and audit semantics, and differ only in which capabilities they can reach.
- **Capability providers**: cloud (sessions, memory, tasks, governed external access), edge executor (local files, shell, Git, browser, private network), MCP (knowledge bases, databases, ticketing, approvals), sandbox (isolated or quota-bound workloads).
- **Cloud brain, local hands.** The edge executor lets an agent work inside your workspace without giving up your local data boundary; when it goes offline, cloud-side memory, sessions, tasks and reports stay available.
- **Side-effect tiers.** Permissions are decided by what a call actually changes — read-only, local change, external read, external change, high risk — not by the tool's name. Planning mode lets an agent think and propose while side effects stay blocked.
- **Three long-term values**: continuity (work pauses, resumes and follows you across devices), control (block one action precisely without killing the task), explainability (which context, which tools, what failed, why this route).

## Solutions

- [Document Intelligence](https://www.matrixorigin.io/solution/document-intelligence): Turn enterprise knowledge into AI-driven productivity.
- [Agentic RAG](https://www.matrixorigin.io/solution/agentic-rag): Evolve RAG from insights into an intelligent executor.
- [LLM Fine-tuning](https://www.matrixorigin.io/solution/llm-finetuning): Build proprietary intelligence from private data.
- [Multimodal Data ETL](https://www.matrixorigin.io/solution/data-etl): Master complex multimodal ETL to unlock data value.
- [AI Data Foundation](https://www.matrixorigin.io/solution/ai-data-foundation): Build next-generation AI data foundations at AI scale.
- [AI Factory](https://www.matrixorigin.io/solution/ai-factory): Enterprise AI factory, co-developed with NVIDIA.

## Resources

- [Documentation](https://docs.matrixorigin.cn/en/): Product docs, guides and API references.
- [Blog](https://www.matrixorigin.io/blog): Engineering, product and research articles.
- [Whitepaper](https://www.matrixorigin.io/whitepaper): Technical whitepapers and solution guides.
- [Case Studies](https://www.matrixorigin.io/cases): Customer stories with measurable business impact across manufacturing, telecom, energy, engineering design and semiconductors.
- [About](https://www.matrixorigin.io/about): Mission, team and company background.

## Code

- [MatrixOne on GitHub](https://github.com/matrixorigin/matrixone): The open-source AI-native database (the "Git for Data" engine).
- [MatrixOrigin on GitHub](https://github.com/matrixorigin): Organization and related open-source projects.