Memory

​

How Memory Works

When memory is enabled on a persona, the platform automatically maintains a memory document for each contact. This document captures facts, preferences, and context from every conversation the contact has with your agents. Before each new conversation, the relevant memory is loaded and injected into the agent’s context. This means your agent can reference past interactions naturally — for example, remembering a customer’s name, their last order, or an issue they mentioned previously.

​

What Gets Remembered

Memory captures factual information from conversations, including:

• Contact details shared during calls (name, preferences, etc.)
• Key topics discussed
• Decisions made or actions taken
• Follow-up items or commitments
• Relationship context with each persona

​

Memory Scope

Memory operates at two levels:

• Global memory — shared across all personas. Facts about the contact that any agent should know (e.g. their name, company, timezone).
• Persona-scoped summary — specific to a single persona. Captures the relationship context between that particular agent and the contact (e.g. “last discussed pricing options on March 5th”).

​

Enabling Memory

Memory is controlled per persona with the memory_enabled setting. When enabled, the platform automatically handles storing and retrieving memory — no additional configuration is needed.

​

How Memory Updates

After each conversation ends, the platform analyzes the conversation and updates the contact’s memory document with any new information. The memory is continuously refined — outdated facts are replaced, and new details are added. This happens automatically as part of the post-call analysis pipeline. You don’t need to configure or trigger memory updates manually.