Skip to main content

​
Overview

The Prompt Block is a single-shot LLM call. You write a prompt, pick a model, run it, get output. It is the simplest and most flexible block, and the workhorse of most workflows. Can receive context from any number of connected upstream blocks. Can also do web search when needed. Perfect for structured tasks, data transformation, and when you need a focused AI response without ongoing conversation. We use it frequently to consolidate “data” after finishing a phase of thinking, preparing for the next phase.

​
Key Features

  • List Processing: Specify if you want output as lists for bulk operations (or a single text response)
  • Model Selection: Choose the best model for your specific task
  • Context Sharing: Automatically shares context with connected blocks (both in and out). Can receive context from any number of connected upstream blocks
  • Web and Doc Search: Specify if you want the prompt block to search through the web and/or all “documents” anywhere on the canvas. Note: This is for broader search — connected blocks will be used as context regardless of these settings
  • Edit in Canvas: Click on the output box to make direct edits to the returned text
  • Re-run Prompt: Click the block, hit the edit button to view and modify the settings used to generate the response

​
How to Use

  1. Create a Prompt Block: Click on canvas → Select “Prompt Block”
  2. Add Your Prompt: Type your instruction or paste your content
  3. Select Model: Choose the most appropriate model for your task
  4. Configure Search: Select web and/or document search (turned off by default)
  5. Select Output Type: Choose text or list format
  6. Run: Click the run button to get your result
  7. Connect for Context: Link to other blocks to use their outputs as input

​
List Processing

When output type is set to list, Prompt Block returns a list and automatically offers two action types:
  • Single Item Processing: Hit the ”+” for just one item to pipe it to the next block
  • Bulk Actions: Apply the same next step to every item in the list, or create a process for one item and apply it across all other items

​
Video Demo