Skip to main content

​
Roles And Content Blocks

In Adaline, messages are the building blocks of your prompts. Each message has two components:
  1. The role: Specifies the actor for the content block
  2. The content block: Defines the actual content of the prompt.
Messages role Choose among the following roles:
  • System: Sets the overall instructions and context of the prompt.
  • User: Represents the end-user inputs.
  • Assistant: Defines the LLM’s output.
  • Tool: Allows you to add tool responses to tool calls in your role-based prompts conversations.

​
Adding Comments

Message comments in Adaline help you annotate your prompts without affecting the actual LLM interaction. They are visible only in the message interface and are stripped out before sending the prompt to the LLM. Use the comment syntax /* Your comment here */ to add notes about prompt design choices or reminders for future edits. Place comments strategically near complex sections to explain your reasoning, making it easier for team members to understand your prompt structure. Comments in Adaline

​
Adding Images

Images bring visual context to your prompts. In Adaline, you can add multiple images per message. Follow the steps below to embed images in your messages:
  1. Navigate to the message block and click on the image button . You can either retrieve the image from a variable, paste the URL of the image, or upload it from your computer: Attaching images in Adaline's messages
  2. Once attached, the system shows the image’s preview: Images preview in Adaline's editor
  3. To add multiple images, click on the plus button and repeat the previous steps.
To learn how to add image variables, read the creating image variables section.

​
Adding PDFs

WARNING: This feature only works with Google’s models.
You can add PDFs to your prompts and ask LLMs to perform actions on them. Follow the steps below to embed PDFs in your messages:
  1. Navigate to the message block and click on the add PDF button . You can either retrieve the PDF from a variable, paste the URL of the image, or upload it from your computer: Attaching PDFs in Adaline's messages
  2. Once attached, you can run your prompt: Run prompts with PDFs
To learn how to add PDF variables, read the creating PDF variables section.

​
Adding Variables

Variables are an integral part of prompts in Adaline. They let you adapt one template to countless situations without rewriting the same content, but customized for specific situations. For example, instead of creating separate prompts for each customer, you can build one powerful template that automatically personalizes based on the variable inputs. This saves you time, ensuring consistency across all interactions.
Variable names must follow specific naming rules. See Variable Name Constraints for detailed information about allowed characters and naming conventions.
The following image shows how to define a variable in the editor: Messages role

​
Dynamic Sources in Variables

Dynamic sources are a key feature that transform static prompts into modular, data-driven workflows. When you use API or prompt in your variables:
  • API Variables allow your prompts to interact with external systems in real-time. You can configure the HTTP method, headers, body, and use placeholders from your dataset columns that are resolved at runtime.
  • Prompt Variables enable prompt chaining, where the output of one prompt serves as input for another. The system automatically handles dependency management and parallel execution.
Dynamic variables can be configured both at the variable level (in the Variable Editor) and at the column level (in datasets), giving you flexibility in how you structure your data-driven workflows. To learn more about setting up variables, read the creating text variables section. For dynamic sources, see API variables and prompt variables.

​
Adding Tool Calls and Responses

There are cases when an LLM can not respond precisely to a prompt for several reasons. Typical reasons are having fresh or specific data. For example, if you ask an LLM what is the weather like in San Francisco right now it will either:
  • Hallucinate the response.
  • Respond, telling it has no access to current data on the weather.
This is one of the many cases where tools come into handy. The Editor allows you to connect LLMs to third party services to expand their knowledge via tools: Tool's fields to fill in in Adaline When working with tools in your prompts, you need to add tool calls and responses to simulate how the LLM interacts with external services. A tool call represents the LLM invoking a specific tool. A tool response shows how the external service will respond back to the LLM. To learn more about tools and how to add and configure them, read the tools section.

​
Creating Role-Based Prompts

Role-based prompting helps your LLM adopt specific personas or expertise. This makes responses more relevant without complex technical setups. Below is a procedure to create effective role-based prompts using a multi-shot prompt:
1

Add a System message

Add a message you want your LLM to respond to.Messages role
2

Upload an image to the User message

Add another message and load an image into itMessages role
3

Add an Assistant message

Add an assistant message to teach the LLM how you would expect it to respond when describing the image.Assistant expected answer
4

Upload another image to the User message

Add another message and load another image into it.Messages role
5

Add another Assistant message

Add another assistant message to teach the LLM how you would expect it to respond when describing the image.Messages role
6

Upload the last image to the User message

This will be the image the LLM will respond to.Messages role
7

Run the prompt in the Playground and see the LLM response

Run the prompt in the playground and verify the LLM response adheres to the ones you added using the Assistant role.LLM's response
TIP: Use the Assistant role in a multi-shot prompt to guide an LLM on how you would expect it to respond to your actual query.