Hysen Labsprompt-optimizer
An AI prompt optimizer for writing better prompts and getting better AI results.
Prompt Optimizer: one tool, four ways to run it
Prompt Optimizer takes a half-finished prompt and pushes it toward sharper AI output. It runs as a web app, a desktop application, a Chrome extension, or a Docker container.
What a round of optimization changes
The pitch is simple: write a prompt, click once, and get a better version back. The README points to one-click optimization with multi-round iterative improvements, and there are two lanes, system prompts and user prompts. To tell whether anything actually improved, it offers analysis, single-result evaluation, and side by side comparison. A list of supported models runs through OpenAI, Gemini, DeepSeek, Grok, Zhipu AI, SiliconFlow, and MiniMax, with model specific parameters set through a field called llmParams.
Three demos that make the point
A minimal role prompt about being a hard-nosed reviewer turns into a structured critique that surfaces weak assumptions and missing evidence. A marketplace bargaining template swaps in item details, price anchors, and tone, so the same prompt works across different conversations. The text-to-image example is the most concrete: a one-line idea, something like a floating library in the night sky, becomes a key visual with clearer subject cues, spatial relationships, and mood anchors. The README frames it as direction, not just expansion.
Where prompts come from and where they go
Prompts start as manual writing, templates, local imports, or Prompt Garden codes. Saved favorites are resource-aware, with version history, reproducible examples, media, and source binding, and can be exported and imported with their referenced resources. An advanced testing mode handles context variables, batch replacement, and multi-turn conversations, plus function calling for OpenAI and Gemini. Image generation covers text-to-image, image-to-image, and multi-image setups, with style transfer from reference images.
Running it yourself
The online version is a pure frontend project, so data never leaves the browser. Forking into Vercel is the recommended web path, and a Cloudflare deploy button exists for quicker setups. The desktop application sidesteps browser CORS entirely and supports automatic updates. Docker Compose spins up the app with an MCP server at the same port, which is how Claude Desktop gets configured. One honest caveat in the README: the macOS build is unsigned, so users will see an unverified developer warning.
Editorial conclusion
The online version keeps every prompt in your browser. For teams that need more, there is a desktop build without CORS limits, a Chrome extension, and Docker with an MCP server for apps like Claude Desktop.
Community notes