Promptimizer

Live website: https://promptimizer.optiqo.dev/ · 📋 Public Roadmap

Paste a rough prompt, get back a structured, model-tailored rewrite — with chain-of-thought, few-shot, and XML technique toggles, per-target-model guidance (Claude, GPT, Gemini, or local), iterative refinement, a word-diff view, and a before/after test run against the raw vs. optimized prompt.

Runs entirely on Cloudflare Pages + Pages Functions. No database, no build step — a static HTML page plus a few small API functions that call whatever OpenAI-compatible, Anthropic-compatible, or Google-compatible endpoint you configure.

---

🔍 How It Works

Promptimizer takes a user's rough prompt and applies advanced prompt engineering principles to rewrite it into a highly structured, model-tailored version. Under the hood, it performs the following steps:

1. Meta-Prompting: Uses optimized system instructions to guide an LLM to critique and restructure your prompt.
2. Technique Toggles: Users can toggle advanced prompting techniques such as:
   • Chain of Thought (CoT): Guides the model to reason step-by-step before outputting the final response.
   • Few-Shot Examples: Inserts dynamic/static placeholder examples to teach the model style/format.
   • XML Formatting: Structures inputs and outputs with XML tags for optimal readability by modern LLMs.
3. Model-Specific Guidance: Tailors the optimized prompt structure depending on the target model selected (e.g., Claude, GPT, Gemini, or Local).
4. Iterative Refinement: Allows users to refine prompts iteratively, analyzing the differences in a clean word-diff view.
5. Comparison Test Bench: Runs a before/after comparison test against both the raw prompt and the optimized prompt in parallel using live completion APIs.

---

🛠️ Getting Started & Hosting Locally

Prerequisites

• Node.js (v18 or higher recommended)
• A Cloudflare account (if deploying to Cloudflare Pages)

Local Setup

1. Clone this repository:

   git clone https://github.com/OptiLabResearch/promptimizer.git
   cd promptimizer

2. Install dependencies:

   npm install

3. Configure Environment Variables: Copy the example environment variable file to .dev.vars and add any provider API keys you wish to use during development:

   cp .dev.vars.example .dev.vars

   Note: .dev.vars is ignored by Git and will never be committed.

4. Start the local development server:

   npm run dev

   This will start wrangler pages dev, serving the frontend and API routes on http://localhost:8788.

Bring Your Own API Key (Default)

By default, anyone using your hosted instance can click the settings gear in the top-right corner to configure their own LLM provider (Groq, Google, OpenRouter, NVIDIA NIM, OpenAI, Anthropic, or any custom OpenAI-compatible endpoint) and API key. These keys are stored only in the user's browser (localStorage) and sent directly to the serverless function per request — they are never logged or stored server-side.

---

🌐 Deploying to Cloudflare Pages

You can deploy the project to Cloudflare Pages in one of two ways:

1. GitHub Git-Integration (Recommended): Connect this repository directly in the Cloudflare Pages dashboard for automatic deployment on every git push.
2. Manual CLI Deployment:

   npm run deploy

---

⚙️ Optional: Hosting a Shared Free-Tier Key

If you want visitors to be able to use the tool without entering their own key, you can configure shared API keys as environment variables/secrets on the Pages project in the Cloudflare Dashboard (Settings → Environment Variables / Secrets):

• HOSTED_PROVIDER: Comma-separated list of providers, tried in order (e.g., groq,google,nvidia,openrouter). If every model on one provider errors out (e.g., rate-limited), the next provider in the list is automatically tried as a fallback.
• HOSTED_MODEL_<PROVIDER>: Comma-separated list of model IDs scoped to that specific provider (e.g. HOSTED_MODEL_GROQ="openai/gpt-oss-120b").
• Secrets (GROQ_API_KEY, GOOGLE_API_KEY, NVIDIA_API_KEY, OPENROUTER_API_KEY): Configure these as encrypted secrets in the Cloudflare Pages settings. Never expose keys in plain environment variables or commit them to the repository.

---

📂 Project Layout

public/index.html                           The single-page frontend application (vanilla HTML/CSS/JS)
functions/api/optimize-prompt.js            POST — one-shot optimize/refine/critique endpoint
functions/api/optimize-prompt/stream.js     POST — SSE streamed optimize endpoint
functions/api/optimize-prompt/test.js       POST — test run comparison endpoint
functions/_lib/optimizer.js                 Shared prompt-building & LLM provider execution logic

---

🤝 Contributing

We welcome contributions to Promptimizer! Here are some ways you can help:

• Optimize Templates: Improve the prompt optimization templates located in functions/_lib/optimizer.js.
• Add Model Guidelines: Update target model-specific rules to match the latest best practices.
• UI/UX Enhancements: Suggest and implement improvements to the frontend user interface in public/index.html.
• Bug Fixes: Submit pull requests for issues found in either the frontend or backend Pages Functions.

Run Tests

Ensure all unit tests pass before submitting a PR:

npm test