Taking how to create the encapsulated agent Chain Operation Analysis as an example, this document focuses on the specific agent encapsulation configuration process.
Choose Admin Console > Agent, click Create Agent, select Encapsulated Agent in the pop-up box, and enter the agent configuration page to complete the following configurations.
Select one or more large language models (LLMs) that have been added under Admin Console > Model from the drop-down list. After adding them, you can select one model in this agent for analysis.
After models are added, you can select Disable Thinking Mode for Chats by Default to disable the thinking mode by default for all chats of the agent, which can reduce unnecessary token consumption and improve response speed in scenarios such as data query.
(1) The model provides the underlying AI inference capability for the agent. Ensure that the model connection is normal.
(2) If the selected model is a model with autonomous thinking, the Disable Thinking Mode for Chats by Default setting will not take effect.
(3) After the Data Agent plugin is upgraded to V1.4, enabling this setting for an existing agent will take effect only for newly created chats and will not change the thinking mode settings of historical chats.
The system prompt is an exclusive instruction that sets the role, rules, capabilities, and output standards at the underlying level of an agent. It is fixed text sent to the agent in advance before you ask the agent a question.
It is equivalent to setting a factory persona and an operating manual for the agent. Users cannot view or modify it, and the agent will strictly follow this set of rules throughout the process to answer all subsequent questions.
Click the + button to select one or more analysis subjects/dashboards that have been added under Admin Console > Data. You can also click Go to Data in the Add Data pop-up box to create analysis subjects/dashboards and then associate them.
In the Add Data pop-up box, click a data source from the left directory to quickly preview the analysis subject and dashboard on the right in the pop-up box.
For preview details, see Data Preview.
Add the knowledge library prepared under Admin Console > Knowledge to the knowledge configuration part of the agent, allowing the agent to rely on this knowledge library for data analysis and report generation.
Click the + button to select one skill that has been created under Admin Console > Skill. The agent will automatically invoke the corresponding capability based on user questions.
After a skill is successfully added, its name and description can be modified to help administrators identify the skill more quickly.
The available data can be adjusted for an added skill to improve its precision and accuracy.
Currently, only the configurations of the following four skills can be adjusted:
FineBI Dashboard Search
FR Report Search
Analysis Subject Data Query
Hard-Coded Orchestration Report
An added skill can be deleted to adjust the agent's capability scope.
Click the + button to select one or more MCP tools added under Admin Console > Open Integration > External MCP, allowing the agent to use external data sources and skill tools.
You can add scheduled tasks for the agent, allowing the agent to automatically execute tasks in the console and push results to target users at specified times via channels such as the Dora platform, WeCom, Telegram, and Teams.
For configuration details, see Scheduled Task.
You can connect the agent to the following IM tools so that the agent can be used within these third-party applications.
For configuration details, see IM Tool Connection.
(1) Multi-Round Chat
This function is enabled by default and cannot be disabled. Context-aware chats are supported for continuous analysis.
(2) Welcome Message
The interactive chat content can be set for the agent's homepage and will be displayed in real time in the right preview area.
(3) Input Box Placeholder
Set the default placeholder text in the input box.
(4) Recommended Question in First Round
Recommended questions can be set for users' first chat to guide user interaction.
For details, see Recommended Question in First Round.
During the configuration, the right Preview area will display the agent homepage effect in real time.
You can view the agent name and welcome message.
You can enter test questions in the input box to verify the agent's response effect.
You can open the workbench to view, reference, and download source files and output files.
You can click Reset Chat to clear the test chat and start a new test.