Prompt Testing
Write, iterate, and save prompts in the Playground using the built-in editor, Handlebars variables, and the prompt library.
Prompt Testing
The Playground gives you a fast feedback loop for writing and refining prompts. You can write messages inline, use template variables for dynamic values, load prompts from the library, and save changes back as new versions.
Writing Prompts Inline
Open the Playground
Navigate to your project and click Playground in the sidebar.
Select a model
Click the model selector at the top of the window and choose a provider (OpenAI, Anthropic, Google, Azure, Bedrock, and others) along with the specific model.
Add messages
Each message block has a role selector (System, User, or Assistant). Click + Message at the bottom of the message list to add more turns. Drag message blocks to reorder them.
Run the prompt
Click Run (or press Cmd+Enter / Ctrl+Enter) to execute. The model response streams in real time in the output panel below.
Continue the conversation
After a response, click Continue in the output panel to append the assistant reply to the conversation and add a new user message — building a multi-turn exchange.
Toggle Stream responses in the settings (⚙) menu to switch between real-time streaming and a single complete response.
Using Variables
Prompts support Handlebars-style variables: {{variableName}}. The Playground detects them automatically and shows an input field for each in the Variables panel on the left.
- Text — enter a static value directly in the Variables panel.
- Image — attach an image URL for vision-capable models.
- Dataset — map the variable to a dataset column for batch execution.
Click the Insert {{input}} button inside a message block to add a variable placeholder at the cursor.
Loading a Prompt from the Library
Open the prompt selector
Click Load Prompt (the prompt icon) in the window toolbar to open the prompt selector.
Pick a prompt and version
Browse or search your prompt library. Select the prompt name and the version (or label) you want to load.
Edit locally
The messages and saved model parameters are populated into the current window. All edits are local to this session until you explicitly save.
Saving Back to the Prompt Library
When you're happy with a revised prompt, use the Save as Prompt button (💾 icon) in the window toolbar:
- Save as new prompt — creates a brand-new prompt in the library with the current messages and model configuration.
- Save as new prompt version — search for an existing chat prompt by name and append a new version to it.
Only chat prompts are supported by the Playground save flow. Text prompts cannot be saved from here.