Sign inGet started

Prompt

Send instructions to LLMs with variable tagging, tool usage, and structured responses

Prompt Cell

Prompt cells let you send instructions to an LLM of your choice and get a structured response back. They’re perfect for natural language queries, text generation, or structured outputs like JSON.

How It Works

  1. Write Your Prompt – Enter your instructions in the User Prompt field.
  2. Choose a Model – Select any available LLM from the Model dropdown.
  3. Configure Advanced Options:
    • Temperature – Controls randomness. Lower values (e.g., 0.1) make the output more deterministic; higher values (e.g., 1.0) make it more creative.
    • Max Tokens – The maximum number of tokens (words/subwords) the model can generate in its response.
    • Top P – Controls nucleus sampling. The model considers the smallest set of words whose cumulative probability exceeds p.
  4. Select Response Format:
    • Text – Returns a plain text result.
    • JSON – Returns a JSON-formatted string. You must describe the expected JSON structure in your prompt.
  5. Optional Fields:
    • System Prompt – Context or behavior guidelines for the model.
    • Tool Usage – Link Python functions (see below) for the model to call.
    • Variable Tagging – Use {{variable_name}} to insert values from other variables in your notebook directly into the prompt.

Variable Tagging Examples

  • Personalizing output: 
    User Prompt: Write a welcome email to {{customer_name}} introducing our new product.
  • Data-driven questions: 
    User Prompt: Analyze the dataframe {{sales_df}} and summarize the top 3 regions by revenue.
  • Dynamic AI chaining: 
    System Prompt: You are an expert analyst.
User Prompt: Using the summary from {{previous_ai_output}}, suggest three follow-up analyses.

Using Tools

You can define a Python function in another Python cell and describe it as a tool:

  • Paste the function in a Python cell — Livedocs will auto-generate the tool call definition in JSON.
  • Or, manually describe the tool’s input/output format in the Tool Usage field.

This lets the LLM call your function to retrieve data or perform computations.

Output Format

  • Text Output Example:

    {
    'result': "The letter 'r' appears 2 times in the word 'strawberry'."
}

  • JSON Output Example:

    {
    'result': '{"count": 2}'
}

    Note: For JSON, the result is stringified.

Next Steps: