LLM Service
Overview
The Guandata BI platform provides intelligent services powered by large language models. Features such as Dashboard Insights and Card Insights use LLMs to deliver deep analysis and visual interpretation of data, bringing users more efficient and accurate analysis and intelligent Q&A experiences.
Before enabling these intelligent services, you must configure an LLM connection to ensure stable availability. The platform supports multiple LLM configurations at the same time to meet the needs of different use cases.
Only users with administrator permissions can configure LLM services.
Steps
- On the Admin Center > System Integration > LLM Service page, click Add Model in the upper-right corner and fill in the model configuration.
- Configure the users or user groups allowed to use the model. Models are available to everyone by default. To restrict access, grant permission to specific users or user groups.
- Test the LLM service connection and save the configuration after the test succeeds.
- Choose the default model. Subsequent intelligent operations automatically use the default model.
- Optional: edit configured models, manage permissions, view lineage, or delete models.
Add a New LLM Configuration
On the Admin Center > System Integration > LLM Service page, click Add Model in the upper-right corner to open the model configuration dialog.
The LLM service configuration currently supports four interface types: OpenAI, Azure OpenAI, Dify, and Guandata LLM Service.
OpenAI
| Configuration Item | Description |
|---|---|
| Display Name | Custom display name for the model configuration, used to distinguish different interface configurations |
| Interface Type | OpenAI |
| Endpoint URL | Fill in the endpoint URL, for example: https://api.deepseek.com/v3 |
| API Key | The key used for authentication and authorization. A valid token is required |
| Model | The model used by the system for default calls. It must match a model supported by the configured endpoint. Different vendors may require different values. For example, Volcengine requires a Model ID, while Alibaba Cloud requires a Model Name. See Additional Notes on the OpenAI Model Field |
Azure OpenAI
| Configuration Item | Description |
|---|---|
| Display Name | Custom display name for the model configuration, used to distinguish different interface configurations |
| Interface Type | Azure OpenAI |
| Endpoint URL | The endpoint URL of the Azure OpenAI resource, in the format https://{resource-name}.openai.azure.com |
| API Key | The key used for authentication and authorization. A valid token is required |
| Model | The deployed model name in Azure OpenAI. It must match the identifier configured in the Azure resource |
| API Version | The Azure OpenAI API version, for example 2025-01-01-preview. It must match a version currently supported by Azure |
Dify
| Configuration Item | Description |
|---|---|
| Display Name | Custom display name for the model configuration, used to distinguish different interface configurations |
| Interface Type | Dify |
| Endpoint URL | API endpoint URL of the Dify application |
| API Key | The key used for authentication and authorization. A valid token is required |
| inputs | Optional. Click Add to append key-value pairs used to pass dynamic parameters required by the Dify application, such as user attributes or dashboard attributes |
Guandata LLM Service
Guandata LLM Service is supported in version 7.4 and later.
-
Enter a display name for the model to distinguish it from other configurations.
-
Click Interface Type, then click Enable beside Guandata LLM Service in the dropdown list. The system automatically completes activation and grants a CNY 100 trial credit.
NoteWhen the trial credit is exhausted, contact your customer success manager to discuss additional credit options if you want to continue using the service.
-
Select the LLM you want to use. Search is supported.
Configure Permissions
Set Initial Access
When adding a model, use Permission to set its initial access scope:
- Available to everyone: all users can select and save the model on editing pages. Both new and existing models are available to everyone by default.
- Available only to yourself: only the current user can select and save the model on editing pages.
Adjust Access
After saving a model, you can change the users or user groups allowed to use it:
- In the model list, click Permission in the model's action column to open Permission Management.
- To restrict access to specific users or user groups, first remove the entry granting access to everyone from the authorization list.
- Select users or user groups, search for and select the recipients of the permission, then click Add. Repeat this step to authorize additional users or groups.
Permission Rules
Default models selected in Set Default Models by Feature are exempt from model permission restrictions and are available to everyone.
Model permissions restrict editing operations only. They apply to the editing pages for Dashboard Insights, Card Insights, and Insight Agent:
- Authorized users can view, select, and save the model on editing pages.
- Unauthorized users cannot select the model. If the current content already uses it, they must switch to a model they are authorized to use before saving the configuration.
For example, suppose the GPT-5.6-terra model used for Dashboard Insights is not configured as a feature's default model, and only one user is included in its authorization list. Only that user can use the model on editing pages; all other users lack permission:
-
If Dashboard Insights does not already use GPT-5.6-terra, unauthorized users do not see it in the model dropdown on the editing page.
-
If Dashboard Insights already uses GPT-5.6-terra, unauthorized users see the model marked as unavailable due to insufficient permission and cannot save the current configuration. They must select another model they are authorized to use before saving.
Test the Connection and Save
After the configuration is complete, click Test Connection to verify that the LLM service can respond normally. Only after the connection test succeeds can you click Save to complete the configuration and use the intelligent services.
Set Default Models by Feature
In the Default Model Settings by Feature section below the model list, choose the default models used by Smart Insights and AI Assistant.
- AI Assistant does not support Dify-type LLMs.
- Default models selected here are exempt from the restrictions configured under Configure Permissions and are available to everyone.
Other Operations
Settings
Click Settings in the corresponding model's operation column to edit its configuration.
If intelligent services behave abnormally later, go to Admin Center > System Integration > LLM Service, click Settings for the relevant model, and run Test Connection again to confirm whether the connection is still valid. If the test fails, update the configuration and retest.
Lineage
Click Lineage to view the resources associated with the model, their Smart Insights switch status, and their paths, such as dashboards and Smart Insight cards.
Delete
Click Delete. The system prompts you to choose a replacement model. After selecting the replacement and clicking Confirm, the model configuration is deleted.
Additional Notes on the OpenAI Model Field
| Model Provider | Model / Model ID |
|---|---|
| OpenAI | For example, GPT-5.1 may support values such as gpt-5.1 and gpt-5.1-2025-11-13. Check the OpenAI official documentation for the latest supported names |
| DeepSeek | Supports only values such as deepseek-chat and deepseek-reasoner. deepseek-chat corresponds to the non-thinking mode of DeepSeek-V3.2, and deepseek-reasoner corresponds to the reasoning mode. Older versions may use values such as deepseek-v3, deepseek-r1, or deepseek-v3.1 |
| Volcengine | For example, deepseek-v3-2-251201. See the Model List for Model ID values |
| Alibaba Cloud Bailian | For example, qwen3-max or qwen3-max-2025-09-23. See the Model List for Model Name values |
| Gemini (OpenAI-Compatible Interface) | For example, Gemini 3 Pro may use gemini-3-pro-preview |
| SiliconFlow | For example, deepseek-ai/DeepSeek-V3.2. See the Model Plaza for available values |
| Zhipu (GLM) | Supports only the GLM series, such as glm-4.6 or glm-4-FlashX-250414. Note that glm must be lowercase. See the Model Overview |
| Moonshot (Kimi) | Supports the K2 series only, such as kimi-k2-0905-preview. See kimi-k2 |