Prompt Audit
Template
Tools
No tools configured.
LLM Audit
No audit results yet.
Assistant & user messages
HTTP Headers
Custom HTTP headers to include in the LLM request. You can use Nunjucks variables like
{{ variable_name }} in header values.
Global variables
(Applied to every test. Test-level values with the same name override these)
Tests & Results
| Test name | Repeat | Pass score | Request time (s) | First response | Actions | |
|---|---|---|---|---|---|---|
| No tests added yet. | ||||||
- https://api.openai.com/v1
- https://openrouter.ai/api/v1
- https://router.huggingface.co/v1
OpenAI API type
- https://generativelanguage.googleapis.com/v1beta
Judge prompt
Judge system prompt
LLM Audit Metaprompt
Prompt Tester is BYOK (Bring Your Own Key) tool that helps you to build and evaluate LLM requests. Think of it as the Postman for LLMs.
Main features:
- One .html file application
- Supports OpenAI (responses, completions) and Gemini protocols
- Template-based (Nunjucks) prompt, messages and tests definitions for flexible testing
- Extracts the prompt and the messages from raw OpenAI and Gemini request payloads
- AI-assisted tests' evaluation (judge)
- LLM Prompt Audit panel to analyze the prompt for issues
Quick Start Guide
- Configure API Settings: Navigate to the Settings tab to enter your OpenAI or Gemini API key and select your preferred model.
- Create LLM Request: In the LLM request tab enter raw LLM request.
If you have your prompt and/or messages included into the payload you can extract them using
Extract prompt & messages button. Use
{{ variable_name }}for dynamic content. - Create the Prompt: In the Developer Prompt tab, enter your prompt. Dynamic content is available via variables defined in the Tests tab.
- Create Messages: In the Messages tab, define the conversation.
- Create Tests: In the Tests tab define your tests. You can specify "Expected" answers for the judge to compare against.
- Run Tests: Once finished, review the results table. Click any row to see the full request and response history.
- Save / Restore Work: Use Save JSON to download your current setup and Load JSON to restore it later.