> ## Documentation Index
> Fetch the complete documentation index at: https://tracevell.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Playground

Playground is an interactive sandbox to experiment with prompts, models, and tools—either from scratch or directly from traced spans.

## What You Can Do

* **Prompt experimentation**: Test prompt variations and inputs with instant results.
* **Open from spans**: Reproduce exact configurations from any LLM span with **Experiment in playground**.
* **Tool integration**: Configure and test custom tools that models can call.
* **Session history**: Review and replay past runs with full context.

![Tracevell Playground Interface](https://mintcdn.com/laminarai/W6ojRY5YjRjfXRin/images/playground/playground.png?fit=max\&auto=format\&n=W6ojRY5YjRjfXRin\&q=85\&s=203fc76b825a440e0bdc363daef95aab)

To access the playground:

1. In the Tracevell dashboard, open **playgrounds** and click **Playground**.
2. Pick a profile and model, then start experimenting, or open directly from a span.

## Models come from LLM profiles

The model picker is built from your workspace's [LLM profiles](https://tracevell.com/docs/platform/llm-profiles.md): one submenu per profile, listing the models that profile exposes. There is no built-in catalog, so a model is available here when a profile lists it and holds credentials for it. A workspace with no profiles shows **No LLM profiles configured** in place of the picker, with a link to workspace settings.

## Start from Scratch

* Create a new playground to iterate quickly.
* Send requests, view responses, and compare outputs side by side.
* Save versions to revisit later.

## Start from a Span

Open the actions menu next to the span name on any LLM span and click **Experiment in playground**. The playground inherits:

* **Model settings** (model, temperature, token limits, thinking tokens)
* **Images** and input formatting
* **Tools** and tool configuration
* **System prompts** and message history

This removes the need to recreate complex requests—tweak and rerun immediately.

![Playground opened from a traced span](https://mintcdn.com/laminarai/W6ojRY5YjRjfXRin/images/playground/playground-from-span.png?fit=max\&auto=format\&n=W6ojRY5YjRjfXRin\&q=85\&s=56f3b1ed957e4a0a6708b819f8ef7d23)

## Tools

Add tools so models can call functions during a session. Tools are defined as JSON with function names, descriptions, and parameter schemas; choose when tools are allowed (`none`, `auto`, `required`, or a specific function).

![Tracevell Playground Tools](https://mintcdn.com/laminarai/W6ojRY5YjRjfXRin/images/playground/tools.png?fit=max\&auto=format\&n=W6ojRY5YjRjfXRin\&q=85\&s=b0feb8531f8059d0388cb6d514f03a45)

**Example tool config**

```json theme={null}
{
  "searchDatabase": {
    "description": "Search for information in the company database",
    "parameters": {
      "type": "object",
      "properties": {
        "query": { "type": "string", "description": "Search query or keywords" },
        "category": {
          "type": "string",
          "enum": ["users", "orders", "products"],
          "description": "Database category to search"
        },
        "limit": {
          "type": "number",
          "description": "Maximum number of results to return",
          "default": 10
        }
      },
      "required": ["query", "category"]
    }
  }
}
```

Learn more about tool choice options in the [AI SDK documentation](https://ai-sdk.dev/docs/ai-sdk-core/tools-and-tool-calling#tool-choice).

## History

Every playground run is saved. The **History** tab shows:

* Full conversations (prompts and responses)
* Model configurations (model, temperature, tokens, etc.)
* Tool calls with inputs/outputs
* Performance metrics (latency, tokens, cost)
* Timestamps for each run

![Tracevell Playground History](https://mintcdn.com/laminarai/W6ojRY5YjRjfXRin/images/playground/history.png?fit=max\&auto=format\&n=W6ojRY5YjRjfXRin\&q=85\&s=faad3b8962c6b7793fa0263075829994)

Open any past run to review, compare, or duplicate it for new experiments.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.