Get started using PromptLayer in minutes.

  • Create and version prompts visually
  • Programmatically retrieve and iterate on prompts
  • Run prompts automatically with OpenAI, Anthropic, and all major models
  • Log and evaulate prompt versions

First, make sure you have signed up for an account at to begin 🍰

PromptLayer is available in Python, Javascript, and REST.

Create your first prompt template

PromptLayer makes creating, versioning, and collaborating on prompt templates easy. They are model-agnostic and auditable.

  1. From PromptLayer’s home screen, navigate to the prompt registry (read more).
  2. Click “Create Template” to create a new prompt template.

Find Registry

For this example, let’s create a new prompt template called “ai-poet”:

You are a skilled poet specializing in haiku. 

Your task is to write a haiku based on a topic provided by the user. 

The haiku must have 17 syllables.
Structured in three lines of 5, 7, and 5 syllables respectively. 

Next, add a new message with an input variable {topic}. This will be filled out by user input when interacting with the AI.

Haiku Prompt

Finally, choose a model to run the prompt on by clicking “Parameters”

Save the prompt template, and we are good to go! 🍰

Set Up PromptLayer Locally

Install PromptLayer & OpenAI

Install the required packages:

Ensure you have the latest versions if already installed.

API Key Env Var Setup

  1. Retrieve your OpenAI API key
  2. Get a PromptLayer API key from settings (click the cog on the top right).
  3. Create a .env file with the following:

For more on setting up environment variables in Python, refer to this guide.

If you are using Python, we recommend using python-dotenv:

Create a Python file called and load the environment variables:

Import PromptLayer

This quickstart uses OpenAI, but PromptLayer supports most major LLM providers, including Anthropic, LLama, Google, Cohere, and Mistral.

First, import PromptLayer and create a PromptLayer client. We’ll use this client to call OpenAI.

Behind the scenes, the PromptLayer library is a wrapper around OpenAI’s SDK. This means that all LLM requests are still made locally from your machine. PromptLayer functions as a sidecar, logging the response after the request is made.

Retrieve & Run the Prompt

Remember the {topic} user input variable we made to specify the haiku topic in our prompt? PromptLayer can format the f-string or Jinja2 template using this variable.

We will use PromptLayer to fetch the latest version of the prompt, inject input variables, and run it.

For simplicity, the code snippet above will work with OpenAI only. Use the prompt blueprint return item for a model-agnostic response format.


Amazing! will automatically format your prompt template into the format needed by OpenAI, Anthropic, and all other major model providers.

The LLM request runs locally on your machine.


In the background, PromptLayer will save the output of every request you make. Open the dashboard to see the logs.

Opening Log

Try opening the log in Playground!

No more redeploys!

Every time you update a prompt in the dashboard, simply re-run the code and PromptLayer will grab the latest version. No eng redeploy is needed.

Redeploy Gif

Decoupling prompts from your code and using a prompt CMS will speed up development. Read more about prompt management best practices on our blog.

For large-scale production deployments, we recommend using webhooks to maintain a caching layer.

Release Labels

PromptLayer allows you to use release labels as a way to manage prod/staging environments.

Back in the PromptLayer dashboard, hover over a version and click “Add Release Label”. Now you can add prompt_release_label to the run call.

Release Label

Alternatively, use prompt_version to specify a version number directly.

Logging Metadata

PromptLayer can be used to monitor user behavior, beta deployments, and much more.

Add a metadata object to each log, tags, and a score to better track requests:

This will help you triage through error logs when using advanced search, creating datasets, or fine-tuning.


Visit the dashboard to see analytics. Learn more

Analytics Screenshot

You can also take advantage of our advanced search by using metadata to search in the sidebar. This is useful for triaging errors by user ID, execution ID, or by score.


Each time you edit the prompt version, you can run an evaluation. Most teams use historical backtests and regression tests to prompt engineer more effectively.

You can build these eval datasets using the request logs we captured above. Learn more here.

Eval Scores

Continue reading with Quickstart Part 2 ➡️

Further reading 📖

From the docs:

Articles on prompt engineering:

Technical tutorials: