Quick start¶
MOI turns enterprise data scattered across systems and formats into knowledge that AI can retrieve, then builds Q&A and automated applications on it. This example follows four stages from ingestion to use: ingestion, processing, knowledge base, and agents.
Example data: A company has two datasets: quarterly sales-detail tables and market-analysis reports (PDF). The numbers are in tables and the analysis in documents, in separate systems that cannot be corroborated. The goal is to put both into one knowledge base, support queries across structured and unstructured data, and produce a quarterly analysis report.
1 · Ingest¶
Upload documents or connect an existing database to import business tables; the latter supports continuous synchronization.
After ingestion, the platform records ownership, source, and relationships in the Catalog. The Catalog is the platform’s metadata directory: it records the existence and relationships of data, not the data itself. Documents and tables are still raw and cannot yet be retrieved by AI.
2 · Process¶
Raw data must be processed before AI can use it: documents must be parsed, chunked, and vectorized, while business tables must be extracted into structures suitable for semantic queries. Workflows perform this processing as pipelines that convert raw data into AI-ready form.
Choose a built-in template or describe the processing goal in natural language for the platform to generate and run a workflow. Documents become vectorized text chunks, tables become semantic structures, and outputs are written back to the database. The Catalog records the processing source of each output—the workflow and steps that generated it—so the path is traceable.
3 · Knowledge base¶
A knowledge base is a dataset open for AI retrieval. Put processed tables and documents in the same knowledge base so one query covers structured and unstructured data. This is a key difference between MOI and single-format solutions.
Queries use natural language; they require neither SQL nor manual document review. For example, when asked why East China declined in the second quarter, a knowledge base retrieves both values from sales tables and discussion from reports, then returns an answer with source citations that you can check item by item.
4 · Agents¶
Q&A provides answers; agents deliver results. When you need a directly usable report or distributable conclusion rather than a paragraph, use an agent.
Agents run on knowledge bases and connect to external tools through MCP. They can turn conclusions from tables and documents into a quarterly analysis report, export it as local HTML, or send it to collaboration platforms such as Feishu or WeCom. This completes the path from raw files to business output.
→ Agents
The data-path cycle¶
An agent’s output is new data that can be ingested, processed, and added to a knowledge base again as the basis for later queries. Data in MOI does not flow through this path only once; it is continuously reused and enriched.
Next steps¶
Try it yourself: Tutorial: turn course materials into a Q&A knowledge base walks through a real dataset to Q&A validation.
Understand the architecture: What is MOI? explains the four layers and how the Catalog connects them.
Browse features: Links at each stage lead to the corresponding feature documentation.