Single prompt

Single prompts in Maxim provide a powerful way to experiment with system and user prompts. This feature allows you to iterate over prompts, test their effectiveness, and ensure they work well before integrating them into more complex workflows for your application.

Working with Single prompts

1. Create a prompt

   Start by creating a new prompt in the Maxim platform.

2. Experiment with sessions

   Use sessions to iterate and save your work as you go.

3. Publish versions

   When satisfied, publish a version of your prompt.

4. Test and evaluate

   Run tests on your prompt versions using datasets and evaluate the results.

5. Deploy

   Deploy your prompt versions with specific conditions using tags.

6. Integrate

   Use the Maxim SDK to access your deployed prompts in your applications.

Creating a prompt

To create a new Single prompt:

Navigate to the "Prompt" tab on the left side panel.

Click on "Prompt" and choose "Single prompt".

Click the "+" icon.

Choose to create a new prompt directly or create a folder for better organization.

Name your new prompt and optionally assign it to an existing folder.

Configuring your prompt

Once you've created your prompt, you can configure it to suit your specific needs:

Selecting a model

Maxim supports a wide range of models, including:

• Open-source models
• Closed models
• Custom models

This flexibility allows you to experiment seamlessly across various model architectures.

Adding system and user prompts

In the prompt editor, you can add your system and user prompts. The system prompt sets the context or instructions for the AI, while the user prompt represents the input you want the AI to respond to.

Configuring parameters

Each prompt has a set of parameters that you can configure to control the behavior of the model. These parameters are different for each model and can be found in the model's documentation. Here are some examples of common parameters:

• Temperature
• Max tokens
• topP
• Logit bias
• Prompt tools (for function calls)
• Custom stop sequences

Using tags

Tags make prompts uniquely identifiable and can be used with the Maxim SDK for targeted prompt retrieval. Below is an example of how to assign tags to a prompt:

Example of using tags with the Maxim SDK:

const prompt = await maxim.getPrompt(
	promptId,
	new QueryBuilder()
		.and()
		.deploymentVar("Environment", "test")
		.tag("CustomerId", 1234)
		.tag("grade", "A")
		.tag("test", true)
		.exactMatch()
		.build(),
);

Working with variables and context

Variables and context are essential components in creating dynamic and adaptable prompts. Let's explore how to use them effectively in Maxim.

Using variables in prompts

Maxim allows you to include variables in your prompts using double curly braces {{ }}. This feature is useful for creating flexible prompts that can adapt to different inputs.

Integrating context sources

You can attach your RAG (Retrieval-Augmented Generation) pipeline to prompts using variables. When Maxim encounters a variable in curly braces with a context source attached, it retrieves the context from the RAG pipeline and replaces the variable with the retrieved context.

Versioning and sessions

Maxim introduces a powerful versioning and session system to help you manage your prompt development process effectively.

Working with sessions

Sessions act as a history, saving the entire state of your prompt as you work. This allows you to experiment freely without fear of losing your progress.

Publishing versions

Versions, as the name suggests, represent different releases of your prompt. Each version has its own system (and optionally a user prompt and description).

When you're ready to test your prompt or deploy it:

1. Click the "Publish version" button.
2. Optionally include the second message (user prompt) in the version and add a description to track changes and the reasoning behind this version.

Testing and evaluating prompts

To thoroughly test and evaluate your prompt:

1. Publish a version of your prompt.
2. Set up a test run with the appropriate evaluators and dataset.
3. Analyze the results to ensure your prompt performs as expected.

For detailed instructions on testing, refer to the Testing a Prompt guide.

Deploying prompts

Once you're satisfied with your prompt's performance, you can deploy it for use in your applications.

To deploy a prompt:

1. Select the version you want to deploy.
2. Click the "Deploy" button.
3. Set the appropriate deployment variables and rules.
4. Confirm the deployment.

Integrating with the Maxim SDK

After deploying your prompts, you can access them in your applications using the Maxim SDK.

const prompt = await maxim.getPrompt("prompt-id");

For more advanced usage, refer to the Maxim SDK Documentation.

By leveraging the single prompt feature in Maxim, you can efficiently develop, test, and deploy prompts for your AI applications. The combination of flexible configuration, versioning, and seamless SDK integration makes it easier than ever to manage your prompt engineering workflow.