> 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/qna.md).

# Q\&A agents

A Q\&A agent answers against exactly one target: a collection or a Brain. Configure the source before testing; the agent does not automatically fall back to another target if its chosen source is unavailable.

## Before you start

You need permission to create or edit agents and access to the intended source. Prepare a synchronized [collection](https://docs.katara.ai/about/documents-and-collections/collections.md), or a [Brain](https://docs.katara.ai/about/brains/overview.md) with usable knowledge. Brain-backed answering also depends on deployment availability; seeing a Brain does not guarantee its query service is enabled.

## Create and configure

1. Open **Agent Library → Create an agent → Q\&A**.
2. Complete the agent details and choose exactly one collection or Brain in the source configuration. Review optional **Context** and **Tags**, then select **Create agent**.
3. Open the saved agent's **Answering** tab and inspect **Answer source**.
4. Confirm **Fixed answer backend**. A collection uses `SOURCE_CORPUS`; a Brain uses `BRAIN_KNOWLEDGE`. This value follows the target and is not a per-question switch. Brain knowledge combines originals and committed Wiki from that one Brain.
5. Under **Answer validation & improvement**, review **Answer validation**, **Minimum score**, and **Answer improvement**. Minimum score is a whole number from 0 through 100; improvement depends on validation being enabled.
6. Under **Caching**, review **Semantic cache**. Select **Save Agent** for a changed draft when you have edit permission.
7. Reopen the configuration to confirm the saved values, then test using [Prompt Lab](https://docs.katara.ai/about/agents-and-answering/prompt-lab.md).

## Understand target immutability

Once a Brain binding is saved, it cannot be changed, cleared, or converted into a collection target, even while the agent is disabled. Create another Q\&A agent for a different target. Do not treat disabling the agent as a reset or try to override the backend in an individual question.

## Verify and recover

A saved agent is configuration, not proof of answer quality or source readiness. Test questions with known source evidence and inspect the returned diagnostics and sources.

If saving is denied, distinguish permission to edit the agent from permission to bind the Brain. If a target is missing or a binding is stale, review current access and source readiness before retrying. Sign in again for an expired session. An incompatible persisted target/backend needs a new correctly configured agent, not an invented default.

For missing evidence, inspect collection synchronization or Brain source/knowledge status. For reusable approved wording and blocked questions, use [Curated answers](https://docs.katara.ai/about/documents-and-collections/curated-answers.md), not the agent's source selector.
