PromptWizard is an open-source prompt optimization framework from Microsoft, available on GitHub with around 4,000 stars. It automates the tedious trial-and-error of prompt engineering by using LLM-driven agents to iteratively refine prompts and few-shot examples together for a specific task.
PromptWizard treats prompt design as a search problem. Given a task description, a small set of examples and a base prompt, the framework runs a feedback loop: it generates candidate prompts, scores them on training data, critiques weaknesses with an LLM, and mutates the best candidates across iterations. Crucially, it optimizes the instruction and the in-context examples jointly, and can also synthesize new training examples when data is scarce. The approach comes from Microsoft Research and has been shown to reach strong accuracy with far fewer model calls than brute-force prompt search. The repository remains actively maintained, with updates through 2025.
PromptWizard is completely free and open source under the MIT license. There is no hosted or paid edition; your only cost is the API usage of whichever models you plug into the optimization loop.
PromptWizard is for LLM application developers, researchers and ML teams who rely on carefully tuned prompts and want a systematic, automated way to squeeze more accuracy out of them.