Supernova — AI Playground

Multi-model prompt evaluation tool used internally at Supernova AI. We built this to make our AI products better. Now we're open sourcing it.

What it does

• Tests prompts across multiple AI models simultaneously via the Vercel AI Gateway (OpenAI, Anthropic, Google, Cerebras, DeepSeek, and many more)
• Shows responses side-by-side for easy comparison
• Streams responses in real-time
• Per-conversation reasoning/thinking effort control (off/low/medium/high) — maps to each provider's native format:
  • OpenAI: reasoningEffort (low/medium/high)
  • Anthropic (Claude 4.6): adaptive thinking (auto)
  • Google: thinkingLevel (low/medium/high)
• Dynamic model discovery from the gateway — no hardcoded model lists
• Stores system prompts and variables separately
• Saves test cases for future testing
• Syncs message deletion across conversations

Why we built this

We were spending too much time testing prompts one model at a time. This slowed down development of our AI tutor. So we built something to test faster.

The interface is intentionally simple. Enter a prompt once, see how different models respond. Save the good ones as test cases.

Why Open Source?

We think more AI companies could benefit from faster prompt testing. And we want others to help make it better.

Running your own instance means full control over data and costs. You only pay for the API calls you make through the Vercel AI Gateway. On Vercel deployments, OIDC authentication is used as a fallback if no API key is set.

Setup

1. Clone and install:

git clone https://github.com/supernova-app/ai-playground supernova-ai-playground
cd supernova-ai-playground
pnpm install

2. Set up environment:

cp .env.example .env
# Edit .env

3. Run migrations (mainly for the authentication):

pnpm db:generate
pnpm db:migrate

4. Start the server:

pnpm dev

Tech Stack

• React + React Router
• TailwindCSS (shadcn/ui)
• Vercel AI SDK v6 + AI Gateway
• Better Auth
• DrizzleORM (PostgreSQL)

License

MIT