Prompt Optimizer is a powerful AI prompt optimization tool that helps you write better AI prompts and improve the quality of AI outputs. It supports four usage methods: web application, desktop application, Chrome extension, and Docker deployment.
Prompts can start from manual writing, templates, local imports, or sources such as Prompt Garden. Prompt Optimizer is where those prompts are optimized, tested, evaluated, and saved as reusable prompt assets.
🎥 Feature Demonstration
Hard-Nosed Reviewer: Turn Agreement into Useful Critique
Starting from a minimal English role prompt, optimization pushes a small model away from generic pushback and toward a clearer, more structured review that surfaces weak assumptions, evidence gaps, and concrete revision advice.
Marketplace Bargaining Reply: Let Variables Change the Strategy
With a single reusable prompt template, you can swap in item details, price anchors, buyer offers, tone, and negotiation goals for different marketplace conversations. After optimization, the same small model does a better job turning those variables into a clearer, more transaction-ready reply instead of a generic helper-style response.
Text-to-Image: Optimize a One-Line Idea into a More Directable Key Visual Prompt
This is not just prompt expansion. Starting from a vague one-line idea, Prompt Optimizer adds clearer subject cues, spatial relationships, and mood anchors. The left side is simply “a floating library in the night sky,” while the optimized version gives the model a more directed fantasy composition that feels closer to a reusable key visual than a lucky generic image.
✨ Core Features
🎯 Intelligent Optimization: One-click prompt optimization with multi-round iterative improvements to enhance AI response accuracy
📝 Dual Mode Optimization: Support for both system prompt optimization and user prompt optimization to meet different usage scenarios
🔄 Analysis and Compare Evaluation: Supports analysis, single-result evaluation, and multi-result compare evaluation to help determine whether a prompt has truly improved
🤖 Multi-model Integration: Support for mainstream AI models including OpenAI, Gemini, DeepSeek, Grok, Zhipu AI, SiliconFlow, MiniMax, etc.
🖼️ Image Generation: Support for Text-to-Image (T2I), Image-to-Image (I2I), and Multi-Image generation with models like Gemini, Seedream, Grok
🌱 Prompt Sources: Start from manual writing, templates, local imports, or Prompt Garden import codes
⭐ Smart Favorites: Resource-aware prompt assets with version history, reproducible examples, media support, source binding, and workspace application
📊 Advanced Testing Mode: Context variable management, multi-turn conversation testing, Function Calling support
🔒 Secure Architecture: Pure client-side processing with direct data interaction with AI service providers, bypassing intermediate servers
📱 Multi-platform Support: Available as web application, desktop application, Chrome extension, and Docker deployment
🔐 Access Control: Password protection feature for secure deployment
🧩 MCP Protocol Support: Supports Model Context Protocol (MCP), enabling integration with MCP-compatible AI applications like Claude Desktop
🚀 Advanced Features
Image Generation Mode
🖼️ Text-to-Image (T2I): Generate images from text prompts
🎨 Image-to-Image (I2I): Transform and optimize images based on local files
🖼️ Multi-Image Generation: Use multiple input images to constrain subject relationships, sequential semantics, and final generation goals
🔌 Multi-model Support: Integrated with mainstream image generation models like Gemini, Seedream, Grok
⚙️ Model Parameters: Support model-specific parameter configuration (size, style, etc.)
📥 Preview & Download: Real-time preview of generated results with download support
🔄 Style Transfer: Learn style, composition, and color from reference images
Prompt Sources & Smart Favorites
🌱 Optional Prompt Sources: Bring prompts from manual writing, templates, local files, or Prompt Garden
📥 Import & Collect: Import prompts with metadata, media, examples, and source binding when available
⭐ Resource-aware Assets: Save stable prompts as reusable favorites with version history
🔗 Source Binding: Track prompt origins and maintain reproducible examples without requiring a specific source
📦 Complete Backup: Export and import favorites with all referenced resources
The online app stores its app data in your browser. Model requests go directly to the provider you configure. For your first run, connect one working text model in Model Manager, then follow the first-time tutorial to improve and copy a prompt.
2. Web Deployment
Vercel Deployment
Method 1: One-click deployment to your own Vercel:
Method 2: Fork the project and import to Vercel (Recommended):
First fork the project to your GitHub account
Then import the project to Vercel
This allows tracking of source project updates for easy syncing of new features and fixes
Configure environment variables:
ACCESS_PASSWORD: Set access password to enable access restriction
VITE_OPENAI_API_KEY etc.: Optional private-deployment model settings. Do not preconfigure API keys on public frontend deployments because VITE_* values are exposed in browser assets.
For more detailed deployment steps and important notes, please check:
Use the Deploy to Cloudflare button for the quickest public-repository setup. It creates a repository under your GitHub/GitLab account and deploys with Workers Builds. For private repositories or stricter repository-access control, import your own repository manually; keep the default deploy commands and clear the build command if Cloudflare auto-fills pnpm run build, because wrangler.jsonc builds the Web frontend and publishes packages/web/dist as static assets.
For access control and analytics on Cloudflare, configure Cloudflare Access and Cloudflare Web Analytics in the Cloudflare dashboard. No frontend dependency or application-code change is required.
3. Download Desktop Application
Download the latest version from GitHub Releases. We provide both installer and archive formats for each platform.
Installer (Recommended): Such as *.exe, *.dmg, *.AppImage, etc. Strongly recommended as it supports automatic updates.
Archive: Such as *.zip. Extract and use, but cannot auto-update.
Core Advantages of Desktop Application:
✅ No CORS Limitations: As a native desktop application, it completely eliminates browser Cross-Origin Resource Sharing (CORS) issues. This means you can directly connect to any AI service provider's API, including locally deployed Ollama or commercial APIs with strict security policies, for the most complete and stable functional experience.
✅ Automatic Updates: Versions installed through installers (like .exe, .dmg) can automatically check and update to the latest version.
✅ Independent Operation: No browser dependency, providing faster response and better performance.
4. Install Chrome Extension
Install from Chrome Web Store (may not be the latest version due to approval delays): Chrome Web Store
Click the icon to open the Prompt Optimizer
5. Docker Deployment
Click to view Docker deployment commands
# Run container (default configuration)
docker run -d -p 8081:80 --restart unless-stopped --name prompt-optimizer linshen/prompt-optimizer
# Run container (with API key configuration and password protection)
docker run -d -p 8081:80 \
-e VITE_OPENAI_API_KEY=your_key \
-e ACCESS_USERNAME=your_username \ # Optional, defaults to "admin"
-e ACCESS_PASSWORD=your_password \ # Set access password
--restart unless-stopped \
--name prompt-optimizer \
linshen/prompt-optimizer
6. Docker Compose Deployment
Click to view Docker Compose deployment steps
# 1. Clone the repository
git clone https://github.com/linshenkx/prompt-optimizer.git
cd prompt-optimizer
# 2. Create .env file for API keys and authentication
cat > .env << EOF
# API Key Configuration
VITE_OPENAI_API_KEY=your_openai_api_key
VITE_GEMINI_API_KEY=your_gemini_api_key
VITE_DEEPSEEK_API_KEY=your_deepseek_api_key
VITE_GROK_API_KEY=your_xai_api_key
VITE_ZHIPU_API_KEY=your_zhipu_api_key
VITE_SILICONFLOW_API_KEY=your_siliconflow_api_key
# Basic Authentication (Password Protection)
ACCESS_USERNAME=your_username # Optional, defaults to "admin"
ACCESS_PASSWORD=your_password # Set access password
EOF
# Because the compose file is under docker/, pass the root .env explicitly.
# 3. Start the service
docker compose --env-file .env -f docker/docker-compose.yml up -d
# 4. View logs
docker compose --env-file .env -f docker/docker-compose.yml logs -f
# 5. Access the service
Web Interface: http://localhost:8081
MCP Server: http://localhost:8081/mcp
You can also directly edit the docker/docker-compose.yml file to customize your configuration:
Click to view docker/docker-compose.yml example
services:
prompt-optimizer:
# Use Docker Hub image
image: linshen/prompt-optimizer:latest
container_name: prompt-optimizer
restart: unless-stopped
ports:
- "8081:80" # Web application port (MCP server accessible via /mcp path)
environment:
- VITE_OPENAI_API_KEY=your_openai_key
- VITE_GEMINI_API_KEY=your_gemini_key
- VITE_GROK_API_KEY=your_xai_key
# Access Control (Optional)
- ACCESS_USERNAME=admin
- ACCESS_PASSWORD=your_password
7. MCP Server Usage Instructions
Click to view MCP Server usage instructions
Prompt Optimizer now supports the Model Context Protocol (MCP), enabling integration with AI applications that support MCP such as Claude Desktop.
When running via Docker, the MCP Server automatically starts and can be accessed via http://ip:port/mcp.
Environment Variable Configuration
MCP Server requires API key configuration to function properly. Main MCP-specific configurations:
In a Docker environment, the MCP Server runs alongside the web application. You can access the MCP service through the same port as the web application at the /mcp path.
For example, if you map the container's port 80 to port 8081 on the host:
In addition to API keys, you can configure advanced LLM parameters for each model individually. These parameters are configured through a field called llmParams, which allows you to specify any parameters supported by the LLM SDK in key-value pairs for fine-grained control over model behavior.
# 1. Clone the project
git clone https://github.com/linshenkx/prompt-optimizer.git
cd prompt-optimizer
# 2. Install dependencies
pnpm install
# 3. Start development server
pnpm dev # Main development command: build core/ui and run web app
pnpm dev:web # Run web app only
pnpm dev:fresh # Complete reset and restart development environment
🗺️ Roadmap
Basic feature development
Web application release
Chrome extension release
Internationalization support
Support for system prompt optimization and user prompt optimization
Desktop application release
MCP service release
Advanced mode: Variable management, context testing, function calling
Image generation: Text-to-Image (T2I) and Image-to-Image (I2I) support
Q1: Why can't I connect to the model service after configuring the API key?
A: Most connection failures are caused by Cross-Origin Resource Sharing (CORS) issues. As this project is a pure frontend application, browsers block direct access to API services from different origins for security reasons. Model services will reject direct requests from browsers if CORS policies are not correctly configured.
Q2: How to solve Ollama connection issues?
A: Ollama fully supports the OpenAI standard interface, just configure the correct CORS policy:
Set environment variable OLLAMA_ORIGINS=* to allow requests from any origin
If issues persist, set OLLAMA_HOST=0.0.0.0:11434 to listen on any IP address
Q3: How to solve CORS issues with commercial APIs (such as Nvidia's DS API, ByteDance's Volcano API)?
A: These platforms typically have strict CORS restrictions. Recommended solutions:
Use Desktop Application (Most Recommended)
Desktop app has no CORS restrictions as a native application
Can directly connect to any API service, including locally deployed models
Provides the most complete and stable feature experience
Use Self-deployed API Proxy Service (Professional solution)
Deploy open-source API aggregation/proxy tools like OneAPI, NewAPI
Configure as custom API endpoint in settings
Request flow: Browser → Proxy service → Model service provider
Full control over security policies and access permissions
Note: All web versions (including online version, Vercel deployment, Docker deployment) are pure frontend applications and subject to browser CORS restrictions. Only the desktop version or using an API proxy service can solve CORS issues.
Q4: I have correctly configured CORS policies for my local model (like Ollama), why can't I still connect using the online version?
A: This is caused by the browser's Mixed Content security policy. For security reasons, browsers block secure HTTPS pages (like the online version) from sending requests to insecure HTTP addresses (like your local Ollama service).
Solutions:
To bypass this limitation, you need to have the application and API under the same protocol (e.g., both HTTP). We recommend the following approaches:
Use the desktop version: Desktop applications have no browser restrictions and are the most stable and reliable way to connect to local models
Use Docker deployment (HTTP): Access via http://localhost:8081, both the app and local Ollama use HTTP
Use Chrome extension: Extensions can bypass some security restrictions in certain situations
macOS Desktop Application Issues
Q5: macOS shows "damaged" or "unverified developer" when opening the app?
A: This is because the application has not been signed with an Apple Developer certificate. Due to the high cost of Apple Developer accounts, the desktop application is currently unsigned.
Solution:
Run the following command in Terminal to remove the quarantine attribute:
# For installed applications
xattr -rd com.apple.quarantine /Applications/PromptOptimizer.app
# For downloaded .dmg files (run before installation)
xattr -rd com.apple.quarantine ~/Downloads/PromptOptimizer-*.dmg
After running the command, you can open the application normally.
🤝 Contributing
Click to view contribution guidelines
Fork the repository
Create a feature branch (git checkout -b feature/AmazingFeature)
Commit your changes (git commit -m 'Add some feature')
Push to the branch (git push origin feature/AmazingFeature)
Open a Pull Request
Tip: When developing with Cursor tool, it is recommended to do the following before committing:
Use the "CodeReview" rule for review
Check according to the review report format:
Overall consistency of changes
Code quality and implementation method
Test coverage
Documentation completeness
Optimize based on review results before submitting
👏 Contributors
Thanks to all the developers who have contributed to this project!
🙏 Acknowledgements
This project was partly inspired by LangGPT in prompt engineering and structured prompt design. Thanks to the LangGPT project and community for their open-source sharing and continued exploration.
In simple terms: You can freely use, modify, and commercialize this project, but if you turn it into a website or service for others, you must share your source code.
👉 Click for detailed explanation
What you can do:
✅ Personal use, learning, and research
✅ Internal company use (not offering public services)
✅ Modify code for commercial projects
✅ Charge for products or services
What you must do:
📖 If distributing software or offering network services, disclose source code
📝 Preserve original author's copyright notices
Core principle: Commercial use is allowed, but not closed-source.
If this project is helpful to you, please consider giving it a Star ⭐️
