Skip to content
Nappai Documentation

Retrieval QA History

This component in Nappai’s dashboard lets you ask questions and get answers based on information from your connected data sources. It remembers previous questions and answers to provide better context and more accurate responses. Think of it as a helpful assistant that learns from your past interactions.

Relationship with Language Models and Retrievers

This component works with a language model (like a sophisticated AI) and a retriever (which finds relevant information from your data). The language model generates the answers, and the retriever provides the information the language model needs to answer your questions accurately. The component combines these technologies to give you the best possible answers.

Inputs

  • System Prompt: This sets the instructions for the AI. It’s pre-filled with helpful defaults, but you can customize it if needed. This is usually best left as is.
  • Input: This is your question. This is the most important input; you must provide a question here.
  • Language Model: This selects the AI model Nappai will use to answer your question. Nappai will guide you on selecting the appropriate model.
  • Retriever: This selects where Nappai will look for information to answer your question (e.g., a specific database, file, etc.). Nappai will guide you on selecting the appropriate retriever.
  • External Memory: This allows you to specify an external source to store and retrieve past conversation history. If left blank, Nappai will use its internal memory.
  • Return Source Documents: Check this box if you want to see the specific data Nappai used to answer your question.
  • Stream: This setting usually remains as the default (True) and manages how the AI processes and returns information. Leave this alone unless you have specific technical reasons to change it.
  • Memory top message: This sets the number of previous messages Nappai remembers for context. The default (20) is usually sufficient.

Outputs

The component doesn’t have explicitly named outputs, but it provides a message containing:

  • Answer: The AI’s response to your question.
  • Source Documents (if selected): If you checked “Return Source Documents,” this shows the data used to create the answer. This helps you understand how Nappai arrived at its response.
  • Conversation History Identifiers: This keeps track of your conversation for future reference.

Usage Example

Let’s say you want to know the total sales for last month from your sales database.

  1. Select your sales database as the Retriever.
  2. Type your question in the Input field: “What were the total sales for last month?”
  3. Click “Run”.

Nappai will use the selected language model and retriever to find the answer in your sales database and display it. If you selected “Return Source Documents,” you’ll also see the relevant data from your database.

Templates

This component is used in many Nappai automation flows. The specific configuration will depend on the flow.

Many components can be used with Retrieval QA History to create complex automation workflows. For example:

  • SQL Query Constructor: Use this to create the queries that your retriever will use.
  • Explain SQL results: Use this to get a human-readable explanation of complex SQL query results.
  • Chat Input: Use this to feed user input directly into the Retrieval QA History component.
  • Chat Memory: Use this to manage and store the conversation history externally. (This is related to the “External Memory” input.)

Tips and Best Practices

  • Clear Questions: Ask clear and concise questions for better results.
  • Relevant Retriever: Make sure to select the correct retriever to access the data needed to answer your question.
  • Experiment with Models: Try different language models to see which one provides the best answers for your data.

Security Considerations

Ensure that the data sources connected to the retriever are appropriately secured and only accessible by authorized users. Review the security settings of your language model and retriever to ensure they align with your organization’s security policies.