// the find
teknium1/Prompt-Engineering-Toolkit
A React app for testing prompts against OpenAI and Anthropic models side-by-side with variable substitution and saved configurations. Targets developers who want a local UI for iterating on prompts without writing scripts. Functionally it's a thin wrapper around two API endpoints.
Side-by-side model comparison is the right UX for prompt iteration — seeing outputs from GPT-4 and Claude in the same view beats switching between chat interfaces. Variable management is a sensible touch; parameterized prompts are how you actually test prompt robustness, not just pasting different inputs manually. MIT license, simple npm setup, no backend required means you can have it running in five minutes.
API keys go in a `.env` file that gets baked into the React bundle — `REACT_APP_` prefixed variables are exposed to the browser, so anyone who opens devtools can read your keys. No diffing or history: once you change a prompt, the previous output is gone, which defeats the purpose of prompt 'engineering'. Only supports OpenAI and Anthropic despite the 'toolkit' branding — no Gemini, no local models via Ollama. Last commit was August 2024, it's effectively unmaintained, and the model lists are almost certainly stale.