Advanced Settings for Deep Dive Analysis

Open

The General section contains settings that control how the AI Space interacts with the configured LLM.

Allow Sending Data to LLM

Use this property to control whether data from the AI Space can be sent to the configured LLM.

When an LLM connection is selected for the AI Space, this setting is validated to determine whether data can be sent to the LLM. If Use Connection Default is selected, the value configured in the underlying LLM connection is used.

The Enable Deep Dive setting unlocks deep dive mode for an AI Space. When enabled, Kyvos Dialogs can plan and reason across multiple tools, and generate grounded, multi-step responses to complex questions.

This section allows you to:

• Enable or disable Deep Dive for an AI Space ( For Admin Users only)

• Configure the LLM used for planning and reasoning.

• Define the maximum number of tool calls the planner can perform for different effort levels.

Open

Controls the LLM connection used by Dialogs to plan and reason in deep dive mode.

• Select the LLM connection from the list. The LLM connection used by Dialogs in deep dive mode for planning and reasoning — determining which tools to call, in what order, and how to interpret results before generating a response.

Controls the maximum number of tool calls that the planner can perform for each effort level.

Set the efforts for tool calls:

• Low Efforts for lightweight tasks. Use this setting for simple queries that require minimal planning and tool usage.

  • Default value: 5

  • Accepted range: 3–7

• Medium Efforts for balanced workloads. Use this setting for queries that require balanced planning and multiple tool invocations.

  • Default value: 10

  • Accepted range: 8–12

• High Efforts for complex, multi-step analysis. Increasing the maximum tool calls allows the planner to perform more reasoning steps and use additional tools when answering complex queries.

  • Default value: 15

  • Accepted range: 13–17

This section describes the LLM and Embedding Connection settings used to power various AI-driven features within the platform. Each feature can be independently configured with its own model connection, allowing administrators to tailor performance, cost, and provider choice per capability.

You can configure separate connections for:

This enables you to optimize each AI capability independently while using the most appropriate model for its specific task.

Open

Configures the models used to convert natural language questions into structured queries.

• LLM Connection: Select the LLM connection used to interpret natural language queries and generate structured queries for retrieving relevant data.

• Embedding Connection: Select the embedding model used to generate embeddings for AI Space metadata. These embeddings enable semantic search by helping the system understand context, identify similar queries, and improve retrieval accuracy.

Configures the model used to generate natural language summaries of query results.

• LLM Connection: Select the LLM connection used to generate meaningful summaries and easy-to-understand explanations from query results.

• Max Data Points for Summary: Specify the maximum number of data points that can be included while generating a summary. This property limits the amount of data processed by the LLM during summary generation. The Default value is 5000

Configures the model used to generate metadata for semantic models.

• LLM Connection: Select the LLM connection used to automatically generate metadata, such as:

  • Descriptions

  • Tags

  • Business context

This information improves metadata quality and enhances searchability and discoverability.

Configures the LLM used by the Web Search tool during Deep Dive analysis. AWS Bedrock does not support Web and URL search.

• LLM Connection: Select the LLM connection used by the Web Search tool to retrieve and synthesize information from web sources during Deep Dive.

Configures the model used for Python code generation and execution during Deep Dive.

• LLM Connection: Select the LLM connection used by the Code Generator tool to write and execute Python code for:

  • Statistical analysis

  • Forecasting

  • Scenario modeling

Configures the model used to compress conversation history.

• LLM Connection: Select the LLM connection used to compact conversation history into concise summaries while preserving important context.

Conversation compaction enables longer Deep Dive conversations by reducing token usage while retaining relevant context.

Configures the model used to summarize retrieved knowledge base content.

• LLM Connection: Select the LLM connection used to summarize Knowledge Base search results into concise, relevant insights before they are incorporated into Deep Dive analysis.