Support Center

Prompting Workbench is a testing platform for AI prompts that helps you validate and improve your prompts systematically. It provides version control, regression testing, and comparison features to ensure your prompts work reliably across different models and scenarios.

To create your first prompt: 1) Navigate to Dashboard > Prompts, 2) Click "Create New Prompt", 3) Enter a name and description for your prompt, 4) Add your prompt content in the text area, 5) Save the prompt. You can then add test inputs and run tests on your prompt.

A prompt is the overall entity that represents your AI instruction, while a version is a specific iteration of that prompt. Each time you modify a prompt, a new version is created, allowing you to track changes over time and compare different iterations.

The number of test inputs depends on your use case complexity. As a general guideline: Simple prompts need 3-5 inputs covering edge cases, moderate complexity prompts benefit from 5-10 inputs, and complex prompts may require 10+ inputs to ensure comprehensive coverage.

Baseline testing involves setting a specific prompt version and its outputs as a reference point. All future changes are compared against this baseline to detect any regressions or unexpected changes in behavior.

Comparison tests run the same prompt with identical inputs across different models or versions. This helps you understand how your prompt performs across various AI providers and identify the best model for your use case.

Yes! Prompting Workbench supports multi-model testing. You can run the same prompt and inputs across different AI models from OpenAI, Anthropic, Google, and Grok to compare their outputs and performance.

Prompting Workbench shows you the outputs from your tests side by side with your baseline (if set). You can review them to determine if the new outputs are better, worse, or equivalent to your expectations, then decide whether to update your baseline or adjust the prompt.

Every time you modify a prompt, Prompting Workbench automatically creates a new version. Each version is timestamped and stores the complete prompt content, allowing you to track changes and compare versions to understand how your prompts evolve.

Yes, you can create alternative version paths to experiment with different approaches. This allows you to maintain multiple variations of a prompt and compare their performance without losing your main development line.

To compare versions: Go to the prompt page, click on "Version History", select two versions you want to compare, and click "Compare". Prompting Workbench will show you a side-by-side diff of the changes.

When you archive a prompt, it is removed from your active workspace but preserved along with all its versions, test inputs, and test results. Archived prompts can be restored if needed in the future.

Prompting Workbench currently focuses on manual testing through the UI. However, you can establish repeatable processes by consistently testing your prompts with the same inputs to track changes over time.

Test results are available within your shared workspace where team members can view and analyze them. Export functionality is coming soon to allow downloading results for external reporting.

A typical workflow: 1) Create your prompt and initial version, 2) Define comprehensive test inputs, 3) Run tests and set a baseline, 4) Make improvements in new versions, 5) Test new versions against the baseline, 6) Update baseline when improvements are confirmed.

Common causes include: No test inputs defined for the prompt, model provider not configured correctly, or invalid prompt format. Check that you have at least one test input and that your AI provider settings are correct.

This could be due to: Using the wrong prompt version, changes in the AI model behavior, or incorrect expected outputs. Verify you are testing the right version and review your baseline configuration.

Version mismatch errors occur when trying to compare incompatible versions. Ensure both versions belong to the same prompt and that they have been properly saved before attempting comparison.

Test performance depends on the AI provider response time and the number of inputs. To improve speed: Run smaller batches of tests, check your internet connection, or contact support if the issue persists.