Database

Overview

The database functionality is a structured data management module provided by the Tuya Developer Platform for agents. It enables agents to perform data storage, retrieval, update, and deletion operations during runtime via natural language or system logic.

Using the database functionality, developers can quickly equip agents with the following capabilities:

• Record user behavior and logs.
• Store personalized profiles and preferences.
• Manage data shared among multiple users.
• Construct business knowledge bases and backend data tables.

The platform provides three data modes to cater to different needs, ranging from personal interaction to collaborative sharing:

• Single-user mode: Each user has an independent data space.
• Multi-user mode: Users share or collaborate on the same dataset.
• Developer mode: Backend data is maintained by developers, while users can only query and use it.

Database modes

Example

The following section demonstrates typical applications of the three modes in real-world agent scenarios.

Single-user mode

Sample interface of a diary card

Multi-user mode

Developer mode

Sample interface of a library AI assistant:

Data import and testing

Procedure

Create database

1. Go to the My Agent page and click Develop in the Operation column of the desired agent.

   

2. Under Model Configuration > Memory, find Database, then click the Add (+) button on the right.

   

3. On the Select Database page, click Add database in the top left corner, or bind an existing database from the list on the right.
   

4. Define the data table structure. Refer to the configuration guide below. Based on the agent’s business type, define the table type, description, fields, and other relevant information.

   

   Item	Configuration description
   Table name	Use a naming format of lowercase letters and underscores (for example, user_diary and book_library). The name should clearly reflect the business purpose and avoid ambiguous or duplicate names.
   Table type	Single-user mode: Suitable for independent data management for a single user, like a personal diary, learning records, and spending logs. Multi-user mode: Supports multi-user collaboration and data sharing, like task leaderboards, message walls, and family ledger. Developer mode: Data tables uniformly managed by the developer, like the knowledge base, product information, and backend FAQs.
   Data table description	Important: This field directly affects the AI’s understanding of the table’s semantics and its matching accuracy. The description should clearly state the purpose and content scope. Example: When users inquire about book-related information (such as title, author, publisher, category, location, and status), prioritize retrieving data from this table for answers and recommendations. This table supports book queries, loan record statistics, reading recommendations, and knowledge retrieval. Providing a complete, specific, and semantic description for each table significantly improves the accuracy of AI invocations.
   Icon	Select a concise and intuitive icon for each table to facilitate quick identification in the console. Icons that are more relevant to the business theme enhance the front-end display and retrieval experience.
   Table field	Each field must include: Field Name: Name in English. Field Description: Function and purpose. Data Type: String, boolean, number, time, and more. Required: Yes/no. Clearer field definitions lead to more accurate AI data retrieval and semantic reasoning. Alternatively, you can generate table fields using AI, then manually adjust or supplement the descriptions in the generated results, as shown in the figure below:

   • The AI will intelligently infer the database structure based on the semantics you provide.
   • More specific descriptions yield more precise structures, including field names, purposes, and data types.
   • Supports multi-turn dialogue for fine-tuning field configurations. The system updates the structure definition in real-time.

5. After completing the form, click Save. The system will automatically initialize the table structure.

   If using Developer Mode, you can click Import in bulk to import sample data for testing.

   

Testing and debugging

Proceed to the Agent Debugging step. You can choose to debug by scanning a QR code or use online debugging to verify the conversation responses.