Skip to content

Prompt Optimizer

Browser-based tool for optimizing, testing and comparing LLM prompts with your own model API keys

System prompt optimization with a rewritten Hard-Nosed Reviewer role and two test outputs side by side

Prompt Optimizer rewrites system and user prompts with a model you choose, then scores or compares the results before you save them as reusable favorites. App data stays in browser storage, and model requests go straight to your provider. On Cloudflare it deploys as Workers Static Assets.

Prompts can start from scratch, a template, a local import or Prompt Garden, and optimization runs over several rounds. Advanced mode adds context variables, multi-turn conversation tests and function calling for OpenAI and Gemini. Image mode handles text-to-image, image-to-image and multi-image generation with models such as Gemini, Seedream and Grok. Favorites keep version history, examples and source bindings, and they export together with their resources. Adapters cover OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter, Ollama, OpenAI-compatible endpoints and several other providers. On Cloudflare the app is a static site. Its wrangler.jsonc builds the core, UI and web packages and serves packages/web/dist through Workers Static Assets, with no server code, database or bindings. The Cloudflare guide advises against building VITE_* API keys into a public site, because they end up in the browser bundle. A desktop app, a Chrome extension and a Docker image are distributed separately.

Built for: Developers, founders and content teams who write LLM prompts and want a private, browser-only optimization tool.

What you can use it for

  • Rewrite a system prompt and compare outputs before and after
  • Test prompts with variables and multi-turn conversations
  • Keep a versioned library of working prompts
  • Refine text-to-image prompts for image models
  • Host a private prompt tool for a team behind Cloudflare Access

Deploy

Deploy with the button, then add model API keys in the app. On a manual import, clear the build command if Cloudflare pre-fills pnpm run build.

What it needs to run

Also uses

  • Anthropic API (optional): One of the model providers for optimizing and testing promptsSource
  • OpenAI API (optional): Model provider for optimizing and testing prompts and for image generationSource

Good to know

The Cloudflare build has no password page, so access control means putting Cloudflare Access in front of it. It also lacks the MCP server, which comes with the Docker image. Browser requests to some model providers fail on CORS, and the README recommends the desktop app as the workaround. Prompts and favorites live in each browser's storage and do not sync automatically. Moving them to another device takes an export or a remote backup to storage you configure.

Cost on Cloudflare

Likely fits the free plan

The deploy serves only static assets, and Cloudflare documents requests to static assets as free and unlimited. Model usage is billed by the AI provider whose key you enter.

Source

The estimate is based on Cloudflare’s documented limits and the app’s configuration. What you pay depends on your usage and plan.

Where this comes from

An alternative to OpenAI Prompt Optimizer, PromptPerfect.

Listing history

  • Oct 2, 2026 · Prompt Optimizer listed
  • VibeSDK home on build.cloudflare.dev with the build prompt box and a grid of community-built apps

    Self-hostable AI app builder from Cloudflare: describe an app and an agent plans, builds, previews and repairs it

    Deploy
    One-click
    License
    MIT
    GitHub stars
    5.4k stars
    Last commit
  • OpenStory new-sequence screen with a script box, aspect ratio picker and style presets such as Action and Neo-Noir

    AI studio that turns a written script into scenes with consistent characters, motion clips and audio

    Deploy
    One-click
    License
    MIT
    GitHub stars
    658 stars
    Last commit