Using Your Assistants

Once your vendor, Knowledge Stores, and Narratives are configured, you can begin interacting with assistants. This is where users ask questions and explore their data in natural language.

An assistant searches only the Knowledge Stores that it has been configured to use. If the user prompting the assistant does not have access to a Knowledge Store, the assistant will not use that data when generating responses.

Custom Assistants

To interact with a custom assistant:

Open the Research Assistant tab.
Click New Chat.
Select the assistant you want to use from the available list.
Enter your question in natural language.

Assistants respond using the Knowledge Stores they are associated with.

Each chat session maintains contextual memory within that session only. Conversations can be saved for later reference.

SQL Assistant

The SQL Assistant is designed specifically for working with structured data. It helps translate natural language questions into SQL queries and allows you to refine or explore those queries interactively. Only assistants configured with the Text-to-SQL agent will be available in the SQL Editor.

To use the SQL Assistant:

Open a new SQL Editor tab.
Click the + new tab icon.
Select the assistant you would like to use.
Enter a question about your structured data.
Review the generated SQL query and execute it as needed.

Each SQL Assistant tab maintains its own session context. Context and history are preserved within a tab, but not shared across different tabs.

You can keep multiple SQL Assistant tabs open to work on different queries simultaneously.

Troubleshooting Assistant Results

If assistants are not returning expected results, review your assistant and Knowledge Store configurations.

Verify that all items in the associated Knowledge Stores have been processed. Only processed documents and data are included in assistant responses.
Review the summarization prompt and ensure it aligns with the types of questions users are asking. Including additional contextual information may improve retrieval accuracy.
Review Knowledge Store contents to ensure that unrelated data has not been included. Mixing unrelated data sources can reduce search accuracy and result quality.

If debugging and Langfuse logging are enabled, you can review execution traces for assistant interactions. These traces show the steps taken to generate a response and can help identify configuration or retrieval issues.