> ## Documentation Index
> Fetch the complete documentation index at: https://docs.honeyhive.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Managing Prompts
> Create, test, and manage prompts in the HoneyHive Playground.
The Playground lets you create and iterate on prompts without writing code. Use it to:
* Experiment with prompt templates and model configurations
* Test prompts against sample inputs before deploying
* Save working versions for use in your application
## Prerequisites
Before using the Playground, configure your model provider API keys in **Settings > AI Provider Secrets**.
You can configure multiple providers (OpenAI, Anthropic, etc.) and switch between them in the Playground.
## Creating a Prompt
1. Navigate to **Studio > Playground** in the sidebar
2. Select a **Provider** and **Model** in the left panel
3. Write your prompt template in the **Chat Template** section
4. Use `{{variable}}` syntax for dynamic inputs (e.g., `{{question}}`)
5. Add sample values in the **Inputs** panel
6. Click **Run** to test the prompt
## Template Variables
Use `{{variable_name}}` syntax in your prompt template to create dynamic inputs. When you add a variable, an input field appears in the **Inputs** panel on the left where you can set sample values for testing.
Variables are replaced with actual values at runtime when you [fetch prompts in your code](/v2/prompts/deploy).
## Hyperparameters
Expand the **Hyperparameters** panel in the left sidebar to configure model parameters:
| Parameter | Description |
| --------------------- | ----------------------------------------------------------------------------- |
| **Temperature** | Controls randomness. Lower values are more deterministic. |
| **Max Tokens** | Maximum number of tokens in the response (UI slider goes up to 4096). |
| **Top P** | Nucleus sampling threshold. |
| **Top K** | Limits token selection to top K candidates. Not available for all providers. |
| **Frequency Penalty** | Reduces repetition of tokens based on frequency. Available for OpenAI models. |
| **Presence Penalty** | Reduces repetition of tokens that have appeared. Available for OpenAI models. |
| **Stop Sequences** | Comma-separated strings that stop generation when encountered. |
For OpenAI reasoning models (o1 series and similar), temperature, top\_p, presence\_penalty, and frequency\_penalty are fixed and cannot be adjusted.
## Response Format
For OpenAI models that support JSON mode, you can set the response format to:
* **Text**: Default. Free-form text response.
* **JSON**: Forces the model to output valid JSON. Useful when you need structured output for downstream processing.
The response format option appears in the Hyperparameters panel when a compatible OpenAI model is selected.
## Multi-Turn Conversations
The Playground supports multi-turn chat. After running a prompt:
1. The model's response appears in the **Conversation** panel
2. Type a follow-up message and click **Run** again
3. The full conversation history is sent with each request
This lets you test how your prompt handles multi-turn interactions before deploying.
## Saving and Forking
Prompts are saved as **configurations** - each configuration is a single record that you can update or fork.
| Action | What Happens |
| -------------------------- | ------------------------------------------------- |
| **Save** (new prompt) | Creates a new configuration with your chosen name |
| **Save** (existing prompt) | Overwrites the existing configuration |
| **Fork** | Creates a copy, preserving the original |
To preserve a working prompt before experimenting, use **Fork** first. Saving an existing configuration overwrites it.
To save a prompt:
1. Click **Save** in the top toolbar
2. Enter a configuration name (e.g., `v1-production`)
3. The saved configuration appears in **Studio > Prompts**
To create a variant without losing the original:
1. Click **Fork** to create a copy
2. Make your changes
3. Save the forked version with a new name
## Managing Saved Prompts
View all saved prompts in **Studio > Prompts**:
From here you can:
* **Deploy** a prompt to an environment (dev, staging, prod)
* **Edit** a prompt by opening it in the Playground
* **Compare** different versions side-by-side
## Opening Prompts from Traces
When debugging production issues, you can open any traced LLM call in the Playground:
1. Go to **Traces** and find the trace
2. Click on a model event
3. Click **Open in Playground** in the top right
This loads the exact prompt template, model, and hyperparameters from that production call so you can iterate on improvements.
## Sharing
To share a prompt with teammates:
1. Save the prompt first
2. Click **Share** in the top right
3. Copy the link
Anyone on your team with access can view and fork the shared prompt.
## Next Steps
Fetch saved prompts in your application via SDK or YAML export.
Test prompt performance systematically with experiments.