Dataset overview-AgentLoop(AgentLoop)-阿里云帮助中心

AgentLoop datasets provide data management capabilities for AI agents and intelligent applications. Datasets serve as a unified layer for business data, knowledge data, and retrievable context. They organize data into structured datasets with customizable fields, support record-level writes and updates, and expose query interfaces that supply agents with real-time, accurate context.

AgentLoop dataset overview

With AgentLoop datasets, developers can consolidate scattered data into standardized data assets. Datasets support common field types such as text, integer, float, and JSON. You can also enable embedding on text or JSON fields, making data available not only for exact filtering and structured queries but also for semantic retrieval, similarity computation, and intelligent recall. For AI applications, a dataset acts as both a data storage container and a retrieval layer that bridges business data with model inference.

Key capabilities

Unified dataset management: Create, view, update, and delete datasets. Resources are organized by AgentSpace, allowing different agents or business domains to manage their own data independently.
Flexible schema definition: Define field types through a schema, describe data semantics, and model internal structures of JSON fields. This adapts to various data shapes including documents, forms, events, and user profiles.
Agent-oriented query interface: A unified query entry point supports data writes, updates, deletes, and reads. Query results include fields, types, row data, and execution metadata, enabling agents to consume data directly.
Retrieval and semantic enhancement: Supports keyword search, conditional filtering, SearchExpr expressions, SQL queries, and embedding-based semantic distance, similarity, and top-k retrieval.
Data lifecycle governance: Dataset metadata and schema changes follow a clear management process. Schema evolution is supported, preventing drift between data definitions and actual query capabilities.
Production-ready service layer: Built-in support for authentication, AgentSpace validation, request tracing, metering logs, and error codes. This makes it straightforward to integrate datasets into real agent application pipelines.

Basic concepts

Dataset: A logical container for a category of business data or knowledge data.
Schema: The structural definition of a dataset. A schema describes field names, types, and retrieval-related configurations.
Field: A data unit within a dataset.