Refrase
your prompts
Better prompts for every model, instantly. No LLM required — it's deterministic, open source, and free.
Click "Refrase It" or select a model to see the optimized prompt.
See the difference
The same prompt, adapted for different models. Each adaptation is backed by tested optimization rules.
You are a helpful assistant that extracts key information from documents. Be thorough and accurate. Use null for any fields not found.
<instructions> You are a helpful assistant that extracts key information from documents. </instructions> <rules> - Be thorough and accurate - Use null for any fields not found </rules>
Quality Score Comparison
+22% improvement
How it works
Three steps. No sign-up required.
Paste your prompt
Drop in any system prompt or user message. Works with plain text, structured prompts, or templates.
Select your model
Choose from 11 model families. Refrase knows the quirks and strengths of each model variant.
Get optimized output
Receive your adapted prompt instantly. Deterministic rules, no LLM needed, under 1ms.
Build prompts with AI
Describe what you need in plain language. The prompt builder generates production-ready prompts with best practices baked in -- guardrails, formatting, and model-specific tuning.
Try the BuilderYou
I need a prompt for extracting contact info from PDFs. It should return structured JSON with name, email, phone, and company.
Refrase Builder
Here's an optimized extraction prompt for your use case: "Extract all contact information from this document. Return a JSON object with the fields: name, email, phone, company. Use null for any field not found. Do not infer or fabricate data." I've included null-handling and anti-hallucination guardrails. Want me to optimize it for a specific model?
Simple pricing
The adapter is free, forever. Pay only for the builder and extra storage.
Pro
- 200 builds / month
- Unlimited saved prompts
- API access
- Priority support
Built on research
Refrase's adaptation rules come from a rigorous evaluation framework. We tested 46 model configurations across 12 model families on structured output tasks -- extraction, analysis, and generation -- to find what actually moves the needle for each model.
Every rule is backed by measurable quality gains. No guesswork, no prompt-engineering folklore.
Read the researchStart optimizing your prompts
Free to use. No account required for the adapter.