Newsletter
Join the Community
Subscribe to our newsletter for the latest news and updates
Automatically create prompts and make them fight each other to know which is the best
Prompts Royale is an open-source, browser-based prompt engineering application that automatically generates prompt candidates, pits them against each other in Monte Carlo matchmaking and ELO rating battles, and ranks them to reveal the best prompt for your task.
Prompts Royale is a prompt engineering tool that turns a task description and example test cases into ranked prompt candidates. It takes a natural language description of what you want a prompt to accomplish, plus a few scenario/expected-output pairs, and generates N candidate prompts that are then battle-tested against each other. The application runs in the browser—the repository is built with Nuxt—and all data stays local; requests go directly from your browser to the LLM API of your choice. The project is maintained by meistrari and is available both as a hosted site (promptsroyale.com) and as a self-hosted Node/Bun application.
Write a prompt that creates a headline for a website, add 2-3 test cases with expected outputs, then let the system generate candidates that fight each other for the #1 spot.bun i to install dependencies, and bun run dev to launch your own private instance.Each prompt is modeled as a normal distribution with an initial mean of 1000 and standard deviation of 350, representing its ELO score. A Monte Carlo sampler picks pairs of prompts to duel; the probability of being drawn is proportional to the chance that the prompt is the best. In each duel, both prompts answer every test case, and a separate judge prompt decides which answer is best. ELO scores are updated with formulas that include a learning rate, K points per battle, and the number of test cases; the standard deviation also shrinks after each match as confidence in the estimate grows.
The source code is publicly available on GitHub under the meistrari account, and you can run it locally for free with Node 16+ and Bun. The hosted version at promptsroyale.com is provided by the developer, but pricing or usage limits are not stated on this page.
It generates multiple prompt candidates from a task description and example test cases, then runs them in automated battles using Monte Carlo matchmaking and ELO ratings. The result is a ranked list of prompts based on how well they perform on your scenarios.
Because requests are made directly from your browser to the LLM API, you need access to an LLM API endpoint. The page doesn't specify which providers are supported, but you'll need to configure your own API access for battles to run.
Yes. In addition to the automatically generated candidates, you can input your own prompt candidates and let them compete against the generated ones.
Node.js version 16 or higher is required to run the project locally. The project uses Bun for dependency management and to run the development server.