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

# Playground

> Admin-only experimentation environment for quick LLM testing — **Chat** and **Completions** modes

The Playground is an **admin-only** experimentation environment for quickly testing LLM models without saving conversation history.
Use for prompt engineering, model comparison, System Instruction validation, etc.

Access directly via `/playground` URL. There's no separate sidebar link.

<Frame caption="Playground Chat mode full screen">
  <img src="https://mintcdn.com/cloocus/jo_Hh1nBizxj0BOc/images/en/chat/playground-chat-full.png?fit=max&auto=format&n=jo_Hh1nBizxj0BOc&q=85&s=6a0e22114b55b4abd972f743208eede6" alt="Playground Chat mode full screen" width="1914" height="908" data-path="images/en/chat/playground-chat-full.png" />
</Frame>

<Warning>
  Only the admin role can access this. Regular users are redirected to home when accessing `/playground`.
</Warning>

<Note>
  All Playground conversations are **ephemeral**. Content disappears when you leave the page; not recorded in chat history or audit logs.
</Note>

***

## Mode Switching

Top tabs offer two modes.

| Mode            | Path                      | Use                                                                        |
| --------------- | ------------------------- | -------------------------------------------------------------------------- |
| **Chat**        | `/playground`             | Multi-turn conversation testing (System Instruction + role-based messages) |
| **Completions** | `/playground/completions` | Single-text completion testing                                             |

***

## Chat Mode

Compose multi-turn conversations to test model responses. Set System Instructions, switch message roles, and edit conversation history.

### Basic Usage

<Steps>
  <Step title="Pick a model">
    From the model selector dropdown in the right **Settings** sidebar (click the gear icon), pick a model to test. The user's default model is auto-selected. If no default is set, manual selection is needed.
  </Step>

  <Step title="Set System Instruction (optional)">
    Expand the **System Instructions** section at the top and enter a system prompt.
    Specify the model's role, constraints, response style, etc.

    <img src="https://mintcdn.com/cloocus/jo_Hh1nBizxj0BOc/images/en/chat/playground-system-instructions.png?fit=max&auto=format&n=jo_Hh1nBizxj0BOc&q=85&s=2fd32e702942e1e0ee61e12e3e2c2839" alt="System Instructions area" width="1913" height="907" data-path="images/en/chat/playground-system-instructions.png" />
  </Step>

  <Step title="Enter message and run">
    Type a message in the bottom input and click **Run**.
    The model generates a response with real-time streaming.
  </Step>
</Steps>

### Message Role Management

In Chat mode, you can freely structure each message's role (User/Assistant).

<Columns cols={2}>
  <Frame caption="User role (1/2)">
    <img src="https://mintcdn.com/cloocus/jo_Hh1nBizxj0BOc/images/en/chat/playground-messages-user.png?fit=max&auto=format&n=jo_Hh1nBizxj0BOc&q=85&s=0c4f7a32affd48baf71f6c4b1856c6bc" alt="Entering a message with the User role" width="1911" height="908" data-path="images/en/chat/playground-messages-user.png" />
  </Frame>

  <Frame caption="Assistant role (2/2)">
    <img src="https://mintcdn.com/cloocus/jo_Hh1nBizxj0BOc/images/en/chat/playground-messages-assistant.png?fit=max&auto=format&n=jo_Hh1nBizxj0BOc&q=85&s=86968ed314a7708f7da94814534bd791" alt="Entering a message with the Assistant role" width="1911" height="907" data-path="images/en/chat/playground-messages-assistant.png" />
  </Frame>
</Columns>

| Feature            | Description                                                                                            |
| ------------------ | ------------------------------------------------------------------------------------------------------ |
| **Role toggle**    | Toggle User ↔ Assistant role with the button on the left of the input                                  |
| **Add**            | Add the message with the selected role to the current conversation without generating a model response |
| **Run**            | Add the current message and generate a model response                                                  |
| **Edit message**   | Click a message to edit its text directly                                                              |
| **Delete message** | Remove with the delete button (minus icon) on the right of the message                                 |

<Tip>
  Useful for **few-shot prompting** tests. Alternate User and Assistant messages to compose an example dialog, then **Run** on the final User message.
</Tip>

### System Instruction

* Collapsible header for efficient screen space
* Shows first-line preview when collapsed
* Edits apply from the next **Run** (no effect on previous responses)

```mermaid theme={null}
sequenceDiagram
    participant U as User
    participant P as Playground
    participant API as Chat Completions API
    U->>P: Enter System Instruction
    U->>P: Add User message (Add)
    U->>P: Add Assistant example (Add)
    U->>P: Enter User question (Run)
    P->>API: messages[system, user, assistant, user]
    API-->>P: Streaming response
    P-->>U: Live text display
```

***

## Completions Mode

Test the model's continuation of a single text input. Focus on pure text generation without conversation structure.

<Frame caption="Completions mode">
  <img src="https://mintcdn.com/cloocus/jo_Hh1nBizxj0BOc/images/en/chat/playground-completions.png?fit=max&auto=format&n=jo_Hh1nBizxj0BOc&q=85&s=3978980c89440f062b98c72157abd99e" alt="Completions mode" width="1911" height="905" data-path="images/en/chat/playground-completions.png" />
</Frame>

<Steps>
  <Step title="Pick a model">
    Click the model name at the top, then pick a model from the dropdown that opens.
  </Step>

  <Step title="Enter starting text">
    Enter the starting text to complete in the text area.
  </Step>

  <Step title="Run">
    Click **Run** — the model generates continuation text in real time.
    Generated text appears appended to the input text.
  </Step>
</Steps>

<Note>
  Completions mode doesn't use **System Instructions**. The input text is passed to the model as-is.
</Note>

***

## Common Features

### Streaming Response

Both modes support real-time streaming via Server-Sent Events (SSE).

* Response is displayed token by token in real time
* Click **Cancel** during generation to stop immediately
* Text area auto-expands to fit content

### Model Selection

* All models enabled in admin settings are available
* Auto-picks the user's default model
* Model name in the dropdown; model ID at the top of the input box

***

## Use Cases

<Accordion title="Prompt Engineering">
  Combine System Instruction and few-shot examples to iteratively test optimal prompts.
  When satisfied, apply the prompt to an [Agent](/en/workspace/agents) or [Prompt](/en/workspace/prompts).
</Accordion>

<Accordion title="Model Performance Comparison">
  Sequentially test multiple models with the same prompt to compare response quality, speed, and cost.
  Chat mode's easy model switching enables fast comparison.
</Accordion>

<Accordion title="System Instruction Validation">
  Pre-test System Instructions before applying to agents.
  Verify consistent responses across various user inputs.
</Accordion>

<Accordion title="Text Completion Testing">
  In Completions mode, check text generation quality for document drafts, code snippets, translations, etc.
</Accordion>

***

## Chat vs. Completions

| Item                    | Chat                          | Completions                        |
| ----------------------- | ----------------------------- | ---------------------------------- |
| System Instruction      | ✓                             | ✗                                  |
| Multi-turn conversation | ✓                             | ✗                                  |
| Role specification      | User / Assistant              | None                               |
| Message edit/delete     | ✓                             | ✗ (edit full text)                 |
| Use                     | Conversational prompt testing | Text completion/generation testing |