Prompt Playground is an interactive, no-code Databricks App for designing, testing, and evaluating prompts stored in the Prompt Registry. It enables product owners, prompt engineers, and both technical and non-technical users to iterate on prompt templates, run them against live model serving endpoints, and evaluate quality at scale — without writing code.
- Manage prompts — browse, create, and version prompt templates directly from the UI
- Iterate interactively — fill in
{{template_variables}}, run against any model serving endpoint, and preview the fully rendered prompt before executing - Evaluate at scale — run a prompt version against any Unity Catalog Delta table, score with built-in LLM-as-judge presets or custom judges, and triage low-scoring results in-app
- Tightly integrated with Databricks — every run and evaluation is logged as an MLflow trace with direct links to the Experiments UI; all data stays in your Unity Catalog environment
- Databricks CLI
>= 0.220.0, authenticated viadatabricks auth login - A SQL Warehouse running in your workspace
- A model serving endpoint — Foundation Model API endpoints work out of the box
- The app's service principal must have
MANAGEon the Unity Catalog schema where your prompts are stored, andMANAGEon the schema where eval results will be written
1. Clone the repository
git clone https://github.com/databricks-solutions/prompt-playground.git
cd prompt-playground2. Deploy
databricks bundle validate
databricks bundle deploy
databricks bundle run prompt_playgroundThe app URL will be printed in the output. You can also find it under Compute > Apps in your workspace.
First time? Start here: Open the app and check out the How to Use tab for a full walkthrough. Click the Settings icon in the upper right to configure your SQL Warehouse, prompt catalog/schema, and evaluation dataset catalog/schema.
Register your first prompt from within the app:
- Open the Prompt Playground app
- Click the + icon next to the Prompt selector
- Fill in a name, optional description, and template (use
{{variable}}placeholders) - Click Create Prompt — the new prompt is registered and immediately selected
App fails to start / "MANAGE privilege" error
→ The service principal is missing MANAGE on the prompts schema.
"No prompts found" → Open Settings (upper right) and verify your prompt catalog and schema are correct. Confirm the app's service principal has access to that schema.
Eval datasets not loading
→ Open Settings and verify your evaluation dataset catalog and schema. Confirm the service principal has MANAGE on that schema.
Model endpoint not listed
→ The endpoint may not be in READY state. Check Serving > Endpoints in your workspace.
For questions or bugs, please contact agents-outreach@databricks.com and the team will reach out shortly.
This project is licensed under the Databricks DB License.
| library | description | license | source |
|---|---|---|---|
| React | Frontend framework | MIT | https://github.com/facebook/react |
| FastAPI | Backend web framework | MIT | https://github.com/tiangolo/fastapi |
| Tailwind CSS | Utility-first CSS | MIT | https://github.com/tailwindlabs/tailwindcss |
| Vite | Frontend build tool | MIT | https://github.com/vitejs/vite |
| MLflow | ML lifecycle management | Apache 2.0 | https://github.com/mlflow/mlflow |
| Lucide React | Icon library | ISC | https://github.com/lucide-icons/lucide |