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

# Redact personal data with a pre-turn flow

> Use a pre-turn flow to mask email addresses and card numbers in customer messages before the agent Klara receives them, and test the result in the Interaction Panel.

> [!NOTE]
>
> **Advanced** · 20 min · Cognigy

Use a pre-turn flow to prepare (or modify) each customer message before agent Klara reads it. In this tutorial, the flow masks personal data.

> [!WARNING]
>
> Pre-turn flows are experimental and off by default. They run only on typed customer messages on supported endpoints. They don't run on voice endpoints (Voice Gateway, AudioCodes, Bandwidth, Voice Copilot, Twilio voice, Avaya, and Alexa), the CXone webhook, or the A2A and MCP servers. On those endpoints, the agent receives the original message. The pre-turn flow changes only what the agent receives. The transcript, project logs, and Insights still store the original message. For details, see [Hooks](https://docs.cognigy.com/ai/agents/develop/hooks.md).

## What a pre-turn flow does

A pre-turn flow is a flow that runs on the customer's message before the agent sees it. The flow can check the message, change it or add information, and then hand the result to the agent. The agent answers the version that the flow hands back, not the original.

The flow uses ordinary nodes, such as **Code**, **HTTP Request**, **If**, and **Lookup**. This lets you add logic that runs the same way every time, without changing the agent's instructions or relying on the model to follow them.

| Use | Example |
| - | - |
| Clean the message | Mask personal data, fix formatting, or translate the message. |
| Add facts the agent should know | Look up the customer's account, check opening hours, or read a CRM record. The flow passes these as *turn context*: labelled notes that the agent reads as background information, not as instructions. |
| Handle button clicks and forms | A click carries data but no text. The flow turns it into a sentence the agent understands, such as "The customer chose: Track parcel". |

The flow hands its result back with the **Continue to AI Agent** node, which has three fields:

* **Text**: the message the agent reads. By default, `{{input.text}}`, the original message unchanged.
* **Data**: structured data passed to the agent.
* **Turn context**: up to 20 labelled entries that the agent receives as background information for this turn.

## What you will build

In this tutorial, you use a pre-turn flow for the first use: cleaning the message. You add a pre-turn flow to Klara, the customer service agent for the fictional company ACME Express. The flow replaces email addresses with `[EMAIL]` and card numbers with `[CARD]`. The masked text replaces the original message in Klara's prompt and conversation history, so the agent never receives the original values. This is true only when the flow runs and hands back its result. See the warning at the top of this page.

The pre-turn flow has the following nodes:

| Node | What it does |
| - | - |
| **Code** | Reads the customer's message, replaces the personal data with placeholders, and stores the result in `context.redactedText`. |
| **Continue to AI Agent** | Passes the masked text to the agent. |

## Before you start

Complete the following prerequisites:

* The project **Klara Tutorial** with the agent Klara. To create them, follow [Build Klara, a customer service agent](https://docs.cognigy.com/tutorials/cognigy-ai/build-klara-customer-service-agent.md).
* Pre-turn flows are available for your organization. If you don't see **Hooks** on the **Advanced** tab, contact your administrator.

## Steps

**1. Find the pre-turn flow setting**

1. In the left-side menu of the project, go to **Design > Agent** and open **Klara**.
2. Go to the **Advanced** tab. Under **Hooks**, find **Pre-Turn Flow**. The description reads "Runs before every agent turn." An agent can have one pre-turn flow.

![The Hooks section on the Advanced tab with the Pre-Turn Flow selector set to Klara pre-Turn Flow and the Open Flow button.](https://mintcdn.com/cognigy-15abf2ba/E7J_TZl63DS6FFRh/_assets/tutorials/pre-turn-flow/hooks-pre-turn-flow.webp?fit=max\&auto=format\&n=E7J_TZl63DS6FFRh\&q=85\&s=f41df83ea9e36b58cd69a5534acdbe0a)

**2. Create the pre-turn flow**

1. Open the **Pre-Turn Flow** list and click **Create new flow**. A message confirms that the pre-turn flow is created.
2. Check that the list shows **Klara pre-Turn Flow**. Cognigy created the flow and selected it for Klara.

> [!NOTE]
>
> Create the flow from this list. A flow that you create on the **Flows** page doesn't appear in the **Pre-Turn Flow** list.

**3. Open the flow**

Click **Open Flow**. The flow chart opens with the following nodes:

* **Start**
* **Continue to AI Agent**, which passes `{{input.text}}` to the agent unchanged
* **End**

At this point, the flow doesn't change the customer's message.

**4. Add the code node**

1. Click the plus icon below the **Start** node. In the search field, enter `Code` and select **Code**.

2. Click the **Code** node. The **Edit Node** panel opens.

3. In the **Code** field, enter the following code:

   ```javascript theme={null}
   const text = String(input.text || "");
   context.redactedText = text.replace(/[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}/g, "[EMAIL]").replace(/\b(?:\d[ -]?){13,16}\b/g, "[CARD]");
   ```

4. Click **Save Node**.

The first line reads the message that the customer sent. The second line replaces each email address with `[EMAIL]` and each card-like number with 13 to 16 digits, with or without spaces or hyphens, with `[CARD]`. Other long numbers, such as phone numbers, are masked too. The result goes to `context.redactedText`.

> [!NOTE]
>
> The patterns are examples. For example, the card pattern also masks phone numbers with 13 or more digits. Before you use the patterns with real customer data, test them with your own data and adjust them.

**5. Pass the masked text to the agent**

1. Click the **Continue to AI Agent** node. In the Node menu, this Node is called **Continue to Agent**.
2. In the **Text** field, replace `{{input.text}}` with `{{context.redactedText}}`. The **Text** field holds the message that the agent receives. If the flow doesn't reach this node, for example because the **Code** node throws an error, Klara doesn't answer the message. The agent never falls back to the original message. For details, see [Hooks](https://docs.cognigy.com/ai/agents/develop/hooks.md).
3. Keep **Data** set to **Forward current input data**. This passes the structured input data to the agent along with the text.
4. Click **Save Node**.

![The Edit Node panel of the Continue to AI Agent node with the text context.redactedText, the data setting Forward current input data, and an empty turn context.](https://mintcdn.com/cognigy-15abf2ba/E7J_TZl63DS6FFRh/_assets/tutorials/pre-turn-flow/continue-to-agent-node.webp?fit=max\&auto=format\&n=E7J_TZl63DS6FFRh\&q=85\&s=e9df4a505bb12b694000261cbde380d3)

![The pre-turn flow with the nodes Start, Code, Continue to AI Agent with the text context.redactedText, and End.](https://mintcdn.com/cognigy-15abf2ba/E7J_TZl63DS6FFRh/_assets/tutorials/pre-turn-flow/pre-turn-flow-chart.webp?fit=max\&auto=format\&n=E7J_TZl63DS6FFRh\&q=85\&s=f8245a296ac2c5ee88c151d2b7c5615a)

**6. Test the pre-turn flow**

1. Go back to the **Advanced** tab of Klara. Check that **Pre-Turn Flow** shows **Klara pre-Turn Flow**.
2. In the upper-right corner, click the chat icon to open the Interaction Panel.
3. Select **Chat**.
4. Enter `My email is jane.doe@example.com. What email address did I just give you?`
5. Read Klara's answer. Klara can answer in different ways, for example by repeating `[EMAIL]` or by saying that it can't repeat the address. The answer alone doesn't prove that the masking worked.

![The Interaction Panel with Klara. The customer message contains an email address and asks which address was given, and Klara answers with the email address \[EMAIL\].](https://mintcdn.com/cognigy-15abf2ba/E7J_TZl63DS6FFRh/_assets/tutorials/pre-turn-flow/test-redaction.webp?fit=max\&auto=format\&n=E7J_TZl63DS6FFRh\&q=85\&s=63ce5514259bd4a1b5a1bbe1164db3d1)

The Interaction Panel shows your original message. The masking applies only to what the agent receives.

To check what the agent received, follow these steps:

1. In the header of the Interaction Panel, click **Advanced mode** (the flask icon).
2. Send the message again. Under Klara's answer, click **Full tracing view** and select the **Pre-Turn Flow** span. The span shows the original text and the final text. In the final text, `[EMAIL]` replaces the email address. This is the text that Klara received.

![The Pre-Turn Flow span in the full tracing view with the original text containing the email address and the final text containing the placeholder \[EMAIL\].](https://mintcdn.com/cognigy-15abf2ba/E7J_TZl63DS6FFRh/_assets/tutorials/pre-turn-flow/trace-pre-turn-flow.webp?fit=max\&auto=format\&n=E7J_TZl63DS6FFRh\&q=85\&s=baffabeb9f98dca5e5892cd82ea3ddc4)

## What you built

You added a pre-turn flow to Klara that masks email addresses and card numbers in customer messages. The **Code** node creates the masked text, and the **Continue to AI Agent** node passes it to the agent. You tested the result in the Interaction Panel.

You can extend the same flow for the other uses: add account details or opening hours as turn context, or turn button clicks into text. An agent has one pre-turn flow, so put all of these steps in the same flow.

## More information

* [Hooks](https://docs.cognigy.com/ai/agents/develop/hooks.md). The Flow itself still records the original message in the project log, the transcript, and Insights. To redact data everywhere, use an Endpoint Input Transformer.
* [Build Klara, a customer service agent](https://docs.cognigy.com/tutorials/cognigy-ai/build-klara-customer-service-agent.md)
* [Hand over between an agent and a flow](https://docs.cognigy.com/tutorials/cognigy-ai/hand-over-between-agent-and-flow.md)