> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agents.labs.bandwidth.com/llms.txt
> Use this file to discover all available pages before exploring further.

# QA

> QA node allows you to run post-call quality analysis on your Voice Agent runs

The QA node lets you define quality analysis criteria that are automatically evaluated after each call ends. Use it to score agent performance, check compliance, or extract structured insights from conversations.

### Creating a QA Node

You can add a QA node to your workflow from the Voice Agent Builder. Once added, configure the analysis criteria that you want to evaluate for each call.

### Choosing the QA model

Leave **Use Workflow's LLM** enabled to use the LLM settings from the workflow version that handled the call, including its model override.

To use a separate model, turn the option off. The provider and model fields use the same catalog as **AI connections**, so catalog updates appear in both places. Select a saved **Credential**, or use **Create credential** to register a named credential. API keys are hidden as you type or paste them. AWS uses its named access-key credentials, while Custom Models lets you enter your own model ID and endpoint.

For direct Claude access, select **Anthropic**, choose a model, and select or create a named Anthropic credential. The [Claude configuration guide](/configurations/llm#anthropic-claude) explains response limits and thinking settings. QA usage includes Claude analysis and summary requests.

Existing QA API keys remain available as **Current saved credential**. Select a credential from the same workspace.

### Editing the QA system prompt

Use **Copy** beside **System Prompt** to copy the instructions. **Expand** opens a larger editor with its own copy control. Closing that editor keeps your changes in the draft; select **Save draft** to persist them.

### Viewing QA Results

After a call completes, the QA analysis runs automatically. You can view the results on the **Run Detail** page for each individual run.