comet-ml--opik/apps/opik-documentation/documentation/fern/docs/prompt_engineering/playground.mdx

---
headline: Prompt Playground | Opik Documentation
og:description: Experiment with LLM prompts in Opik's playground to evaluate performance
  and enhance your prompt engineering skills.
og:site_name: Opik Documentation
og:title: Explore Prompt Playground - Opik
title: Prompt Playground
canonical-url: https://www.comet.com/docs/opik/development/prompt-playground
---

<Note>
  In Opik 2.0, prompts are project-scoped. Playground conversations are logged to the `playground` project by default.
</Note>

<Tip>
  The Opik prompt playground is currently in public preview, if you have any feedback or suggestions, please [let us
  know](https://github.com/comet-ml/opik/pulls).
</Tip>

When working with LLMs, there are times when you want to quickly try out different prompts and see how they perform. Opik's
prompt playground is a great way to do just that.

## Using the prompt playground

The prompt playground is a simple interface that allows you to enter prompts and see the output of the LLM. It allows you
to enter system, user and assistant messages and see the output of the LLM in real time.

You can also easily evaluate how different models impact the prompt by duplicating a prompt and changing either the
model or the model parameters.

<Frame>
  <img src="/img/prompt_engineering/playground.png" />
</Frame>

_All of the conversations from the playground are logged to the `playground` project so that you can easily refer back to them later._

## Configuring the prompt playground

The playground supports the following LLM providers:

- OpenAI
- Anthropic
- OpenRouter
- Gemini
- Vertex AI
- Azure OpenAI
- Amazon Bedrock
- LM Studio (coming soon)
- vLLM / Ollama / any other OpenAI API-compliant provider

<Tip>
  If you would like us to support additional LLM providers, please let us know by opening an issue on
  [GitHub](https://github.com/comet-ml/opik/issues).
</Tip>

Go to [configuring AI Providers](/v1/administration/workspace-settings/ai_providers) to learn how to configure the prompt playground.

## Running experiments in the playground

You can evaluate prompts in the playground by using variables in the prompts using the `{{variable}}` syntax. You can then connect a dataset and run the prompts on each dataset item. This allows both technical and non-technical users to evaluate prompts quickly and easily.

![playground evaluation](/img/evaluation/playground_evaluation.gif)

When using datasets in the playground, you need to ensure the prompt contains variables in the mustache syntax (`{{variable}}`) that align with the columns in the dataset. For example if the dataset contains a column named `user_question` you need to ensure the prompt contains `{{user_question}}`.

Once you are ready to run the experiment, simply select a dataset next to the run button and click on the `Run` button. You will then be able to see the LLM outputs for each sample in the dataset.

### Accessing nested JSON dataset fields

If a dataset column contains JSON-formatted content, you can use dot notation to reference nested values directly when querying or filtering. Dot notation lets you specify the path to a nested field (e.g., `{{input.user.name}}`) to extract only the value you need from within a structured object. This makes it easier to work with deeply nested data without manual flattening or custom parsing.

## Using images in the playground

The playground supports multimodal prompts with images when using vision-capable models. You can add images in two ways:

### Adding images directly in messages

You can add images directly to your prompt messages through the playground UI:

<Frame>
  <img src="/img/prompt_engineering/playground_add_image.png" />
</Frame>

<Note>
  **Internal representation**: When you add an image through the UI, Opik internally stores it with `<<<image>>><</image>>>` wrapper tags on a new line in the prompt. This internal format is not visible in the UI but ensures proper serialization and processing of multimodal content.
</Note>

### Using images from datasets

When evaluating prompts with datasets, you can reference image columns using the standard Mustache syntax:

```
Analyze this product image: {{product_image}}
```

Where `product_image` is a column in your dataset containing image data.

**Supported image formats:**
- Image URL
- Base64 encoded image

<Tip>
  When using images in prompts, ensure you select a vision-capable model. 
  Opik automatically detects which models support vision based on provider capabilities.
</Tip>