By default, AI Agent conversations are stateless, meaning the AI Agent does not retain information from previous customer interactions. Each new conversation starts without awareness of past requests, preferences, or resolved issues.

Customer Memory enables AI Agents to capture and reuse relevant information from previous conversations, allowing them to provide more personalised and context-aware responses without requiring customers to repeat information.

Customer Memory includes:

• Semantic Memory: Long-term customer facts and preferences derived from previous conversations.
• Episodic Memory: Summaries of important events and outcomes from past conversations.

You can configure Customer Memory under Build > Conversation Settings and review generated memory records under Monitor > Customer Memory.

How Customer Memory Works

When Customer Memory is enabled, the AI Agent automatically generates and stores memory from completed conversations.

The process works as follows:

1. A customer interacts with the AI Agent.
2. Once the conversation meets the configured memory criteria and remains inactive for the defined duration, memory generation begins.

3. The AI Agent extracts:

   • Configured profile-level attributes
   • Semantic Memory (facts and preferences)
   • Episodic Memory (conversation events and summaries)
4. The extracted information is stored in the customer's memory profile.
5. During future interactions, relevant memory is automatically added to the AI Agent prompt, enabling personalised responses based on previous customer interactions.

Configure Customer Memory

To enable Customer Memory:

1. Navigate to AI Agent Builder > Build > Conversation Settings > Contextual Memory.
2. Turn on the Enable Contextual Memory toggle.

For additional configuration details, see Conversation Settings: Advanced Controls.

Types of Customer Memory

Semantic Memory

Semantic Memory stores persistent customer facts and preferences that remain relevant across multiple interactions. These memories are continuously updated as new information becomes available.

Examples:

• Customer has a dog named Milo.
• Customer is a Premium customer.
• Customer prefers communication in English.
• Customer recently purchased a digital sketching tablet.
• Customer is moving to a new apartment.

Semantic memories are categorised (for example, Product Usage, Account Characteristics, Pet Information, and Location Changes) and assigned a relevance score.

Episodic Memory

Episodic Memory records specific events from previous conversations in chronological order, helping the AI Agent understand a customer's interaction history.

Examples:

• Customer updated an order delivery address.
• Customer requested an order cancellation.
• Customer submitted a refund request.
• Customer reported a shipping issue.

Unlike Semantic Memory, which captures long-term information, Episodic Memory focuses on individual conversation outcomes and historical events.

View Customer Memory

To review Customer Memory records:

1. Open an AI Agent and select Monitor from the left navigation pane.
2. Open the Customer Memory window.
3. Select the Eye icon next to a memory record to view the customer's profile memory.

Each customer profile contains the following sections:

Customer Information

Displays customer details such as name, contact information, and associated channel profile.

Profile Attributes

Displays configured customer attributes sourced from profile custom fields, such as:

• Customer Tier
• Preferred Language
• Region
• Relationship Manager
• Other custom business attributes

Semantic Memory

Displays customer facts and preferences organised into categories such as:

• Location Changes
• Product Usage
• Account Characteristics
• Pet Information

Each memory entry includes a Relevance Score to indicate its importance for future interactions.

Episodic Memory

Displays chronological customer interaction events, including:

• Timestamp
• Event summary
• Relevance Score

This enables administrators to review significant customer interactions preserved for future context.

Memory Injection During Task Execution

When an AI Agent executes a task, Customer Memory is automatically added to the prompt sent to the Large Language Model (LLM).

The injected context can include:

• Configured profile attributes
• Relevant Semantic Memory records
• Relevant Episodic Memory records

Providing this context allows the AI Agent to generate responses that reflect previous customer interactions, preferences, and history without requiring customers to repeat information.

To view the complete memory injection prompt, select the Injection Prompt icon in the Profile Memory window.

View Memory in Audit Logs

You can verify injected customer memory through Monitor > Audit Logs.

When a task is executed, the audit log displays the complete prompt sent to the LLM, including any automatically injected memory.

This helps you:

• Identify which memories were used during the interaction.
• Review how memory was incorporated into the prompt.
• Validate that the injected context aligns with the AI Agent’s response.

Reviewing audit logs can help troubleshoot memory-related behaviour and ensure that the AI Agent uses the appropriate customer context during task execution.