Evolve your AI SDK prompts right in your codebase.

A playground that edits prompts directly in your codebase, and runs them against real app context.

cool.prompt.ts
import { prompts } from "@evalution/vercel-ai-sdk";

export default prompts(
  { id: "my-cool-prompts" },
  ({ openai }) => ({

    weatherAgent: (query: string) => ({
      model: openai("gpt-5.5"),
      system: `You answer questions about the weather. ` +
              `Today is ${new Date().toDateString()}.`,
      messages: [{ role: "user", content: `Question: ${query}` }],
      temperature: 0.7,
    }),

}));

Live, in your browser

  • Prompts as code. Your prompts live as typed .prompt.ts files — diff, review, and ship them like everything else.
  • Real app context. Run prompts against your actual application data, not throwaway inputs.
  • Swap models live. OpenAI, Anthropic, Google — switch providers and tune parameters without touching code.
  • Built on the AI SDK. Works with the Vercel AI SDK prompts and tools you already ship.
  • Local-first, zero setup. npx evalution and you're running. No account, no keys to paste.

Booting the demo…

Coming Soon

Production traces, dataset storage, team sharing, and synced local-first environments. Drop your email for updates.