DolphinX Web

Starting from version 3.00.6, DolphinDB Web Intereface supports DolphinX-based AI coding. AI coding introduces a new interaction model for developing DolphinDB scripts. After configuring the AI model, you can interact with DolphinDB using natural language to generate, edit, and execute DolphinDB scripts. The system can plan tasks, write the scripts, and complete the work based on your instructions.

Note:
DolphinX AI coding is available only on standalone deployment or on compute nodes of cluster deployment.

1. Initialize and Configure Agent

After deploying DolphinDB 3.00.6 or later, open the web interface to initialize the built-in AI resources and configure the agent.

1.1 Initialize Built-In Resources

  1. Go to http://<IP>:port/ and sign in to the web interface of standalone deployment or compute nodes of cluster deployment as the admin.
  2. Open Shell and click Create AI Session on the right side of the tab bar.
    Figure 1. Figure 1-1 Create AI Session
  3. Click Initialize.
    Figure 2. Figure 1-2 Initialize
  4. After initialization, you can view the default DolphinDB Coding Agent and built-in skills.
    Figure 3. Figure 1-3 DolphinDB Coding Agent
    Figure 4. Figure 1-4 Built-In Skills

1.2 Configure the AI Model

  1. Click AI Settings and then click Model Management.
    Figure 5. Figure 1-5 Provider Settings
  2. Click New Provider, enter the provider name, Base URL, and API Key, select the adapter type, and then click Create.
    Figure 6. Figure 1-6 Create a New Provider
  3. Click New Model. Configure the model details based on its capabilities, and then click Save.
    Figure 7. Figure 1-7 Add a New Model
    Figure 8. Figure 1-8 Newly Added Model

1.3 Bind the Model to the Agent

After adding a model, bind it to an agent for use.

  1. Click Agent Management.
    Figure 9. Figure 1-9 Agent Management
  2. On the Model Binding tab, find the target model and click the bind icon.
    Figure 10. Figure 1-10 Bind a Model
  3. After binding, the agent can use the bound model to write code and execute tasks.
    Figure 11. Figure 1-11 Model Bound

2. Permissions

For the AI coding feature, administrators and standard users have different permissions.

2.1 Administrators

Administrators can manage all configurations for the AI coding module and manage permissions for standard users. Administrators have access to the following features:

  • Agent Management
  • Provider Settings
  • Skill Management
  • Memory Management
  • MCP Management
  • User and Group Management

2.2 Standard Users

Standard users' access to Agents, models, skills, and other resources is granted directly by an administrator or via the user group they belong to. Standard users can perform the following operations:

  • Create AI Session
  • View Session History
  • View and Manage Personal Memory
  • View Personal Usage

  • Configure Token Budget

2.3 Standard Users with Agent Management Permission

Administrators can grant standard users or groups permission to manage agents. Standard users with agent management permission can create and configure agents and manage the models, skills, and MCP servers bound to those agents and knowledge base. For detailed instructions, see User Management.

2.4 Group Permissions

Users can be assigned to user groups so that permissions can be managed in bulk by adjusting group permissions. Administrators can configure Agent permissions, user model permissions, user skill permissions, and Agent management permissions for a user group. Members of a group automatically inherit the permission settings of the group to which they belong. For more information about user group management, see Permission and User Access Control.

3. AI Coding

  1. After configuring an AI model and binding it to an agent, click Create AI Session to use AI coding. You can describe your requirements in natural language and use the agent to complete them. Based on the goals you have set, the agent can break the work into steps and generate DolphinDB scripts.
    Figure 12. Figure 3-1 Create an AI Session
  2. The following example shows how to use it to assist in creating databases and tables. In the chat box, enter “Create a stock database and table using the recommended partitioning scheme of DolphinDB” and then click Send. The agent automatically plans the work, breaks the task into steps, and searches for relevant information.
    Figure 13. Figure 3-2 Start a Conversation
  3. When the agent needs to run the DolphinDB script, you can review and approve the script. After approval, the AI-generated script is executed. If you do not want approvals to interrupt your work, you can choose Allow and remember for this session.
    Note:
    AI-generated scripts may have errors, so approve with caution. If you choose to allow by default for this session, you assume the risk of possible AI errors by yourself.
    Figure 14. Figure 3-3 Allow Agent to Execute the Generated Scripts
  4. The agent completes the task step by step and then provides a summary.
    Figure 15. Figure 3-4 Execution Result
  5. Starting from version 3.00.6.1, even though DolphinDB script execution is allowed by default in this session, the AI assistant will still refuse to perform destructive operations — such as deleting, removing, cleaning up, or undefining — unless they have been reviewed and explicitly approved by a human.
    Figure 16. Figure 3-5 Refuse to Execute Destructive Operations
  6. AI chat supports uploading images and files.
    Figure 17. Figure 3-6 Upload Images and Files
    You can drag and drop images and files into the chat box to upload them.
    Figure 18. Figure 3-7 Drag to Upload
    Note:
    The file upload feature relies on the ModelInference and DocParser plugins. Please ensure these two plugins are loaded before use.

4. Skill Configuration

A skill is a skill package that enhances the agent's ability to handle specific work. Configuring skills can improve the agent's performance. You can configure skills in AI Settings.

  1. Click Skill Management to view the built-in skills.
    Figure 19. Figure 4-1 View Built-in Skills
  2. Administrators can register and manage skills. Click Register Skill to register a new skill.
    Figure 20. Figure 4-2 Register a Skill
  3. Upload a skill package that complies with the specification. For details, see Skill Package Specification.
    Figure 21. Figure 4-3 Upload a Skill Package
  4. After uploading, you can modify the configurations to customize the skill. Click Register to register the new skill.
    Figure 22. Figure 4-4 Customize Skill Information
  5. After registration, the new skill is in the skill list.
    Figure 23. Figure 4-5 Registered Skill
  6. After registering a skill, bind it to an agent for use. Click Agent Management, select the agent to bind the skill to, and then click Skill Binding.
    Figure 24. Figure 4-6 Skill Binding
  7. Find the skill you want to bind, and then click the enable button.
    Figure 25. Figure 4-7 Enable the Skill

5. Personal Center

In Personal Center, you can view the conversation history and token usage. On the Personal Sessions page, you can view your conversation history and perform operations such as renaming titles, generating summaries, manually triggering memory extraction, and deleting sessions.

Figure 26. Figure 5-1 Personal Center

On the Personal Usage page, you can view the current user's token used for API calls.

Figure 27. Figure 5-2 Token Usage

You can view detailed session information through Context Preview.

Figure 28. Figure 5-3 Context Preview

6. User Management

6.1 User

Administrators can manage permissions for standard users, including access to agents, models, and skills. Administrators can also grant standard users permission to manage agents. For each agent, model, and skill, you can control user access permissions by clicking Authorize or Revoke.

Figure 29. Figure 6-1 User Permissions

Administrators can grant standard users agent management permission.

Figure 30. Figure 6-2 Grant Agent Management Permission

Standard users with agent management permission can manage agents.

Figure 31. Figure 6-3 Agent Management

6.2 Group

With user group management, administrators can configure access to Agents, models, and skills for a group of users, without the need to configure permissions for each user individually.

Figure 32. Figure 6-4 Group Permissions

6.3 User Usage

On the User Usage page, administrators can view users' token usage. Usage is broken down by model. Click + to expand the entry and view each user's usage for each model.

Figure 33. Figure 6-5 View Usage of Each User

Click Budget Configuration. Administrators can enable a token budget limit for the selected Agent, as well as set an overall token limit and limit for each model.

Figure 34. Figure 6-6 Token Budget Configuration
Note:
Token usage and budget are available only when the provider returns token usage. If the provider does not return token usage, both features are unavailable.

7. MCP Management

Administrators can configure MCP Servers on MCP Management to register tools that models can call.

Figure 35. Figure 7-1 MCP Management
  1. Click Add Server.
    Figure 36. Figure 7-2 Add a MCP Server
  2. After entering the required information, click Save Configuration.
    Figure 37. Figure 7-3 Enter the MCP Information
  3. In Agent Management, bind the MCP server to the agent.
    Figure 38. Figure 7-4 MCP Bindings

After binding, the agent can call the MCP tools.

8. Memory Management

Memory management is used to manage useful information recorded by the model during its work. In Memory Management, you can manage personal memories and control whether the memory feature is enabled. Administrators can manage public memories.

Figure 39. Figure 8-1 Memory Management

In Memory Management, administrators can configure memory policies for agents.

Figure 40. Figure 8-2 Configure Memory Policies

In Task, administrators can manage memory generation tasks.

Figure 41. Figure 8-3 Manage Memory Generation Tasks

9. Knowledge Base Management

A knowledge base is used to store and manage domain-specific documents. When performing tasks, an agent can retrieve content from the knowledge base as a basis for answering questions and generating scripts, thereby improving the agent's performance in specific business scenarios. Administrators, or standard users with agent management permissions, can configure knowledge bases.

Note:
The knowledge base feature relies on the ModelInference and DocParser plugins. Please ensure these two plugins are loaded before use.
Figure 42. Figure 9-1 Load Necessary Plugins
  1. Click + to create a new knowledge base.
    Figure 43. Figure 9-2 Create a New Knowledge Base
  2. Fill in the required information, then click Create.
    Figure 44. Figure 9-3 Fill in Knowledge Base Configurations

    Once created, the new knowledge base is displayed.

  3. Click Upload document.
    Figure 45. Figure 9-4 Document Upload
  4. Click the file upload box, or drag files into it, to upload.
    Figure 46. Figure 9-5 Start Uploading Documents

  5. Click Start Upload. You can view the parsing status of the files.

    Figure 47. Figure 9-6 View the Parsing Status