AI Playground

The AI Playground is a lightweight interface for testing prompts against OpenAI and Anthropic models directly from your browser. You bring your own API key, choose a model, write a system prompt and user message, and see the live streaming response — without leaving the page.

Your API key is saved only in your browser's localStorage — it never touches our database. When you click Run, the key is sent in the request body to our proxy server, which immediately forwards it to OpenAI or Anthropic and streams the response back. The key is never logged, cached, or stored after the request completes. You can verify this by checking that the key field is empty in a private browsing window where localStorage is cleared.

Anthropic's API does not allow direct browser-to-API calls (CORS policy). To support both providers consistently with proper streaming, requests are forwarded through a lightweight Next.js API route that acts as a passthrough proxy. It reads your request, forwards it verbatim to the provider, and streams the response back — no data is modified or retained.

Temperature controls how creative or random the model's responses are. A value near 0 makes outputs very predictable and deterministic — good for structured tasks like JSON extraction or code generation. Higher values (1.0–2.0) make responses more varied and creative — better for brainstorming or creative writing. Most tasks work well at the default value of 0.7.

A system prompt is a set of instructions sent to the model before the user's message. It defines the model's persona, constraints, and behaviour — for example, "You are a code reviewer who gives concise, direct feedback." The system prompt is optional but dramatically improves output quality for specific tasks.

Yes. While the model is generating, the Run button changes to a Stop button. Clicking Stop immediately aborts the request and keeps whatever response has already streamed in. This is useful when the model starts going in the wrong direction or you want to save API costs.

Costs depend entirely on which model you use and how many tokens your prompt and response consume. GPT-4o mini and Claude Haiku are the most affordable options at well under $0.001 per typical interaction. Use the Prompt Cost Calculator on WiseCrewAI to estimate costs before running long prompts.