AI Studio¶

Positioning¶

AI Studio is a workspace where you connect external data, process it into searchable and queryable AI-Ready data, and then use that data to build agents that can converse and execute tasks. Data engineers process data and agent developers build agents using the same underlying data in the same workspace, without moving data back and forth between two toolsets.

One primary path¶

AI Studio’s features are not a set of parallel menus. They are organized around one primary path from data to agents:

Connectors (bring in external data)
   → Catalog (land raw data)
   → Workflows (parse / clean / chunk / vectorize / extract)
   → Catalog (write processed results back)
   → Knowledge bases (wrap results as searchable datasets)
   → Agents (conversational Q&A, NL2SQL, and task execution)

Every part of this path—including tables and volumes in Catalog, SQL queries, and vector indexes—runs on the MatrixOne database engine underneath. This is explained further under “Handoff” at the end of this page.

Workspaces: the boundary for all resources¶

A workspace is the top-level isolation unit in AI Studio. Connectors, Catalog, compute resources, workflows, knowledge bases, agents, and all other resources belong to a workspace, and data is completely isolated between workspaces. Use the switcher on the left of the top bar to switch between “My Workspaces” and “Joined Workspaces.” Workspace IDs are also used for cross-workspace data sharing and API calls.

Modules¶

Data foundation: Catalog and compute resources¶

Module	What it is	Learn more
Catalog	The workspace’s unified data-asset center. Its object tree is “directory → database → volume / table / model / operator.” Browse, manage, and trace the lineage of loaded raw data and workflow results here	Platform administration · Catalog
Compute resources	Compute capacity for processing and queries. It is strictly divided into two non-interchangeable types: query resources run only SQL, while task resources run only non-SQL processing such as parsing and embedding. Resources are billed by node online time and can scale to 0 when idle	Platform administration · Compute resources

Data in and out: Data Connections¶

Connectors store the connectivity and read/write capabilities of external data sources. Load tasks then materialize data into Catalog, while export tasks send processed results back to external systems. Many source types are supported, including databases, object storage, APIs, and enterprise applications. See Data Connections.

Data transformation: Data Processing¶

Module	What it is
Workflows	Arrange multiple processing steps into a savable, reproducible, and schedulable DAG. The saved DSL is the source of truth for execution; the canvas is only its rendered representation, and output from the AI orchestration assistant is only a candidate awaiting confirmation. Nodes cover data reads/writes, multimodal parsing, cleaning and chunking, embedding, structured extraction, SQL processing, flow control, and more
SQL editor	Interactively browse databases and tables, validate data, write analytical SQL, and save workbooks

Module

What it is

Workflows

Arrange multiple processing steps into a savable, reproducible, and schedulable DAG. The saved DSL is the source of truth for execution; the canvas is only its rendered representation, and output from the AI orchestration assistant is only a candidate awaiting confirmation. Nodes cover data reads/writes, multimodal parsing, cleaning and chunking, embedding, structured extraction, SQL processing, flow control, and more

SQL editor

Interactively browse databases and tables, validate data, write analytical SQL, and save workbooks

See Data Processing.

Make data queryable: Knowledge Bases¶

A knowledge base is a logical wrapper over Catalog, not a separate copy of the data. It hides Catalog path complexity from agents and serves as a collaboration interface between data engineers and agent developers. File-based material goes through parsing and chunking—text, images, tables, and transcripts—while structured data uses semantic configuration, including ten categories such as dimensions, metrics, and table relationships, to support NL2SQL. A knowledge base’s built-in conversation page is a validation layer, not an external publishing layer. See Knowledge Bases.

Put it to work: Agents¶

An agent is an AI capability with bound knowledge bases, skills, tools, and prompts. It can provide conversational Q&A (RAG), query and analyze data (NL2SQL), execute scheduled or triggered tasks, and be published to messaging channels. See Agents.

Cross-cutting capabilities: monitoring, permissions, and billing¶

Module	What it is	Learn more
Monitoring and alerts	Jobs, SQL history, operation logs, and alerts—execution records and exception notifications only within an AI Studio workspace	Monitoring and alerts
User permissions	Users, roles, and feature/object permissions within a workspace; separate from MatrixOne SQL privileges	User permissions
Cost and billing	Credit-based billing for compute resources, AI services, storage, and more	Cost and billing

Data mode and Agent mode¶

The modules above are presented through two modes in the top bar for two types of users. They are two views of the same underlying data:

	Data mode	Agent mode
Intended for	Data engineers	Agent developers / business users
Primary work	Connectors, Catalog, workflows, SQL, compute resources, and permissions	Conversations, binding knowledge bases / skills / tools, and configuring tasks
Interaction	Forms and configuration for professional control	Conversation-driven, with less exposure to underlying details

The intersection of the two modes is the knowledge base. Data mode handles data engineering and semantic quality on one side; Agent mode binds the knowledge base to agents on the other. Both use the same underlying Catalog data.

Handoff: the foundation underneath and the models it needs¶

AI Studio is where data is processed and used, but it does not store data itself or produce models. Those responsibilities belong to the other two platforms, which is how the three platforms connect:

Data storage and compute are provided by MatrixOne. Tables and volumes created in Catalog, queries run by the SQL editor and query compute resources, and knowledge-base vector indexes all run on the MatrixOne hyper-converged database engine underneath. Users generally do not see the engine directly, but when you need to create instances, write SQL, or perform engine-side operations, you enter MatrixOne.
Model supply is connected to Genesis. Agent conversations, knowledge-base Embedding, and workflow model nodes all call models. Within Studio, these capabilities use the platform’s unified “AI services” pricing and can hide the underlying model. When you need an API Key to call models directly from your own application or provide model services externally, you use Genesis, which has a separate ledger. For the boundary between the two, see How the three platforms work together · Ledger separation.