ChatBI Toolset

ChatBI (Data Analysis) Toolset Seamlessly integrates with Xpert Data Analysis Platform for business intelligence tools.

Configuring ChatBI Toolset

See Data Analysis Toolset.

Session Variables

• chatbi_models: List of available Cubes, accessible via get_available_cubes tool.
• chatbi_cubes: List of available Cubes.
• chatbi_cubes_context: Contextual info for available Cubes.
• tool_chatbi_prompts_default: Default prompts for ChatBI tools, usable in main agent prompts.

Prompt Example

You are a sales analysis expert. Use tools to answer user queries.  
{{tool_chatbi_prompts_default}}  

Common times:  
{{sys.common_times}}  

Available Cubes:  
<cubes>  
{{chatbi_models}}  
</cubes>  

Tools

1. get_available_cubes

Fetches list of accessible Cubes. Use {{chatbi_models}} in prompts to reduce tool calls.

Available Cubes:  
<cubes>  
{{chatbi_models}}  
</cubes>  

2. get_cube_context

Retrieves Cube context (dimensions, measures, metrics). Helps agents understand Cube data structure for accurate responses.
When using Session Summary, context may be summarized. Use {{chatbi_cubes_context}} in prompts for detailed Cube info:

Cube context:  
<cubes>  
{{chatbi_cubes_context}}  
</cubes>  

3. dimension_member_retriever

Fetches dimension member details (dimension, hierarchy, level, key, caption) for building reports or filters.
Solves inefficient manual searches by using semantic queries and embeddings to return matching members.

Features:

• Natural language query support.
• Handles multi-level dimension structures.
• Controls TopK results.
• Supports re-embedding for accurate matches.
• Links with model ID and Cube for consistent results.

Use Cases:

• Confirming “East China” for sales analysis.
• Retrieving members like “Gold Card” for filters.

Parameters:
| Parameter | Type | Description |
|-----------------|----------------|------------------------------|
| modelId | string | Data model ID. |
| cube | string | Cube name. |
| query | string | Keyword for fuzzy search. |
| dimension | string (optional) | Dimension name. |
| hierarchy | string (optional) | Hierarchy name. |
| level | string (optional) | Dimension level. |
| topK | number (optional) | Max results returned. |
| re_embedding | boolean (optional) | Re-embed members (default: false). |

Indexing Dimension Members:
Vectorize dimension members for faster retrieval. See Dimension Member Retrieval.

4. answer_question

Answers user queries using Cube context, returning data visuals or metrics.

Temporary Calculated Members:
For queries requiring temporary calculations, agents pass calculated_members to the tool:

{  
  "calculated_members": [  
    {  
      "name": "Calculated Member Name",  
      "caption": "Display Name",  
      "description": "Description",  
      "formula": "Expression",  
      "formatting": {  
        "unit": "Unit",  
        "decimal": "Decimals"  
      }  
    }  
  ]  
}  

Example: Analyze monthly sales growth rate for value-added resellers over the past two years.

1. Fetch Cube context.
2. Retrieve dimension member details.
3. Build visual answer with monthly growth rate calculated member.

Open the chart execution interpreter to view detailed parameters:

Explanation of calculated member parameters

Parameter Control:
For models with parameters, agents select appropriate values.
Example: “Top 5 high-value customer sales” uses:

{  
  "parameters": [  
    {  
      "name": "TopN",  
      "value": 5  
    }  
  ],  
  "measures": [  
    {  
      "name": "High-Value Customer Sales"  
    }  
  ]  
}  

5. show_indicators

Displays metrics as a list when users explicitly request specific indicators.

Feishu-Adapted ChatBI

You are a data analysis expert. Use tools to answer user queries.  

If no specific query, use `welcome` tool to suggest 3 questions for Top 3 models based on context.  

If Cube dimensions/measures are missing, use `get_cube_context` first.  

Use `answer_question` for analysis results:  
- Use `slicers` for explicit time.  
- Use `timeSlicers` for relative time:  
  <input>last year</input>  
  <output>lookBack: 1, lookAhead: -1</output>  
  <input>this year</input>  
  <output>lookBack: 0, lookAhead: 0</output>  
  <input>next year</input>  
  <output>lookBack: -1, lookAhead: 1</output>  
  <input>last year to this year</input>  
  <output>lookBack: 1, lookAhead: 0</output>  

Current date: {{sys.date}}  
Common times: {{sys.common_times}}  

If Cube lacks required measures/indicators, create one using `create_indicator` before answering.  
MDX query restrictions:  
1. Same dimension cannot appear in multiple axes (columns, rows, slicers). Use functions like `descendants` for restrictions.  

No image generation needed; visuals are shown via tools.  

Available models:  
<cubes>  
{{chatbi_models}}  
</cubes>  

Note: show_indicators is not supported for Feishu.

welcome

When no specific query is provided, suggests questions for users based on available models.
Set as a terminal tool; no further response needed after execution.

Publishing to Feishu Bot

Publish ChatBI toolkit as a Feishu bot app via one-click publishing. Displays results as interactive card messages.

See ChatBI Feishu Bot for details.

Version Control & Changelog

• v3.4:
  • Removed create_indicator tool; replaced with calculated_members parameter.
  • Added parameters and calculated_members to answer_question tool.