> For the complete documentation index, see [llms.txt](https://docs.katara.ai/about/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.katara.ai/about/agents-and-answering/prompt-lab.md).

# Prompt Lab

Use **Prompt Lab** to test a configured Q\&A agent with realistic questions and inspect the response. It is a testing surface for the selected agent, not a way to choose a different collection or Brain for each question.

## Before you start

You need answer-generation permission, an accessible Q\&A agent, and a workspace that passes the current access and subscription checks. If **Prompt Lab** is not visible, check [Access](https://docs.katara.ai/about/organization-administration/access.md) rather than assuming that every organization role can use it.

Configure the agent's source and answering settings first using [Q\&A agents](https://docs.katara.ai/about/agents-and-answering/qna.md). For collection-backed agents, check synchronization. For Brain-backed agents, check knowledge readiness and deployment query availability.

## Send a test question

1. Open **Prompt Lab** from the sidebar.
2. Explicitly select the intended Q\&A agent in the composer. The interface can initially select the first loaded agent, so check the name before sending sensitive or source-specific questions.
3. Enter a question of at least 10 characters. Start with a question whose answer you can verify in the selected source.
4. Select **Send message**, or press Enter. Use Shift+Enter to insert a line break.
5. Wait for the answer, or cancel the pending response with the available cancellation control.
6. Read the answer and any displayed source references, confidence information, and diagnostics. Compare the result with the underlying evidence rather than treating a score as a factual guarantee.

The expected outcome is a response from the selected agent or an actionable error. A successful response does not establish that every uploaded file is indexed or that a stale Wiki has refreshed.

## Compare changes

Keep a short set of representative questions, including ones the source cannot answer. Record the selected agent and observations outside the page if you need a durable test record: this interface does not promise persisted chat history.

Change the agent's **Answering** settings, save them, then repeat the same questions. Consider **Semantic cache** and [curated questions](https://docs.katara.ai/about/documents-and-collections/curated-answers.md) when interpreting repeated wording; do not assume every answer was freshly generated.

## Recover from problems

For validation errors, lengthen the question and confirm an agent is selected. If an agent disappears, recheck organization and current access. For a source error or poor answer, inspect the bound collection or Brain before changing prompts. A saved Brain binding is immutable; create a different Q\&A agent to test another target. If the service is unavailable, consult [Availability](https://docs.katara.ai/about/reference/availability.md) rather than repeatedly submitting the same question.
