Skip to content
Nappai Documentation

Tool Calling Agent

This component acts as a smart assistant within Nappai. It uses a language model (like a sophisticated chatbot) and other tools to process your requests and provide helpful responses. Think of it as a central hub that connects your prompts to the right tools to get the job done.

Relationship with Language Models

This component works by interacting with a chosen language model. You select the language model you want to use, and the Tool Calling Agent uses it to understand your requests and generate responses. The better the language model, the more accurate and helpful the responses will be.

Inputs

  • Language Model: This is the brain of the operation. You must select a language model for the agent to use. Nappai will provide a list of available options.
  • System Prompt: This sets the overall instructions for the agent. It defines the agent’s personality and how it should behave. A default helpful assistant prompt is provided. You can customize this to fine-tune the agent’s behavior.
  • Prompt: This is your actual request or question. It needs to include the word “input” to work correctly (e.g., “input: What’s the weather like today?”). A default “{input}” is provided.
  • Chat History: This allows the agent to remember previous interactions within the current conversation. This helps it provide more relevant and contextual responses.
  • Conversational responses: This setting determines whether the agent should respond in a conversational style, remembering past interactions. The default is off (False). Turning it on (True) makes the interaction feel more like a natural conversation.

Outputs

The component produces a text-based response. This response is generated by the agent after processing your prompt and using any relevant tools. The output will appear in the dashboard.

Usage Example

Let’s say you want to find the latest news about the Eurocup and then summarize it. You would:

  1. Select a suitable Language Model.
  2. (Optional) Adjust the System Prompt if needed.
  3. Enter your prompt: “input: Find the latest news about the Eurocup 2024 and then summarize it.”
  4. (Optional) Enable “Conversational responses” for a more natural interaction.

The Tool Calling Agent will then use its connected tools (like a web search tool) to find the news, summarize it, and display the summary as the output.

Templates

This component is used in the following Nappai templates: Eurocup 2024, Chat with GMail, Search Tools. The specific configuration within each template may vary.

The Tool Calling Agent works with many other components in Nappai. Here are a few examples:

  • Google Generative AI Embeddings: Helps the agent understand the meaning of words and phrases.
  • Explain SQL results: Helps interpret data from databases.
  • SQL Query Constructor: Helps build database queries.
  • DALL·E Image Generator: Can generate images based on text descriptions.
  • Many more: The Tool Calling Agent can connect to a wide variety of components to perform different tasks. Explore the Nappai dashboard to see the full list.

Tips and Best Practices

  • Choose the right Language Model: Select a language model appropriate for your task. Some models are better at certain tasks than others.
  • Use clear and concise prompts: The clearer your prompt, the better the agent will understand your request.
  • Experiment with the System Prompt: Adjusting the system prompt can significantly influence the agent’s behavior and the quality of its responses.
  • Leverage Chat History: For conversational tasks, enabling chat history allows for more context-aware and natural-sounding responses.

Security Considerations

Always be mindful of the data you provide to the language model and the tools it uses. Avoid entering sensitive or confidential information unless absolutely necessary. Nappai employs security measures to protect your data, but it’s crucial to use responsible practices.