Prompt Refine

What is Prompt Refine?

Prompt Refine is a tool for systematically improving the prompts you use with language models (LLMs). You can generate and manage prompt variations, run experiments, track performance, and compare results against previous attempts. Variables let you isolate exactly how specific changes affect a model’s responses.

It supports models from OpenAI, Anthropic, Together, and Cohere, plus local models for more customization. Completed runs can be exported as CSV for deeper analysis.

Key features:

• Experiment storage — runs are saved so you can compare them side-by-side and track differences from your last attempt.
• Broad model compatibility — works with multiple AI providers as well as local models.
• Folders — organize your experiment history and switch between prompt tests.
• Prompt versioning — a system inspired by Chip Huyen’s ideas, for tracking prompt performance, exploring variations, and seeing how small tweaks lead to different outcomes.

The beta version allows up to 10 runs.

Who created Prompt Refine?

Prompt Refine was created by @marissamary. A team plan is available, which includes 10 million tokens per month, access for up to 20 people, and email support. For questions, you can reach the creator on Twitter at @marissamary.

What is Prompt Refine used for?

Typical uses include:

• Summarizing large documents with source citations.
• Rewriting text to match a specific style or tone.
• Answering questions based on information within a given text.
• Providing feedback on writing.
• Brainstorming ideas and thinking through considerations.
• Producing initial drafts for reports and presentations.

Who is Prompt Refine for?

Anyone working with LLMs who wants better, more consistent results — whether that means summarizing documents, rewriting content, drafting reports, or testing prompt variations.