Long-term memory usage examples

Apply Unreal long-term memory patterns for returning learners, shared devices, first sessions, and fresh-start reset flows.

Use these patterns after Long-term memory quick start is working. Each example focuses on which identity values to set on the chatbot component before StartSession. For assignment steps, see End-user identity.

Returning learner in a training simulation

A safety training simulation should remember that a learner has already completed a valve inspection module.

Setup:

• LTM is enabled for the training character.
• The learner has a saved SpeakerID.

Blueprint flow:

1. Load SpeakerID from the learner's SaveGame or account profile.
2. Set UConvaiChatbotComponent.EndUserID and UConvaiPlayerComponent.EndUserID to the saved value.
3. Call StartSession on the chatbot component.

Expected outcome: The character can refer to facts learned in earlier sessions for the same learner and character.

First-time player profile

A new player starts the application with no stored identity.

Setup:

• The project has collected a display name, such as Alex.
• No saved SpeakerID exists for the profile.

Blueprint flow:

1. Call Convai Create Speaker ID with Speaker Name set to the display name.
2. On On Success, save the returned SpeakerID.
3. Set both components' EndUserID to the saved value.
4. Call StartSession.

For the full create workflow, see Speaker ID management.

Expected outcome: Convai creates a distinct memory scope for this player and character pair. Future launches should reuse the saved SpeakerID.

Shared classroom device

Several learners use the same machine during guided practice. Device fallback would merge their memory, so each learner needs an explicit identity.

Setup:

• Each learner selects or signs into a profile before interacting with the character.
• Each profile stores its own SpeakerID.

Blueprint flow:

1. On profile selection, load that profile's SpeakerID.
2. Set the chatbot and player EndUserID values before StartSession.
3. Keep each profile's SpeakerID separate from other profiles on the same device.

Expected outcome: The character remembers each learner separately, even though all learners use the same device.

Account-backed identity

An enterprise onboarding app already has a user account ID and does not need a Speaker ID record.

Setup:

• The account system provides a stable user ID, such as employee-1042.
• The account ID does not change across devices.

Blueprint flow:

1. Load the signed-in user's account ID.
2. Set UConvaiChatbotComponent.EndUserID and UConvaiPlayerComponent.EndUserID to the account ID.
3. Optionally set EndUserMetadata on both components with the same JSON string:

{"name": "Jordan Kim", "department": "Facilities"}

1. Call StartSession.

Expected outcome: Memory follows the account identity instead of the local device.

Fresh start for a new scenario attempt

A learner repeats an assessment and should begin a fresh conversation, while keeping the same long-term identity.

Setup:

• The learner keeps the same SpeakerID or account ID.
• The project should not resume the previous conversation session.

Blueprint flow:

Blueprint

Call Reset Conversation on the Convai Chatbot component, then call StartSession.

C++

Expected outcome: The next connection starts from a fresh local conversation link. The player's identity remains the same, so character-level long-term memory can continue to accumulate for that user.

Remove a test identity

During development, remove test identities created by repeated Speaker ID experiments.

Blueprint flow:

1. Call Convai List Speaker IDs.
2. Find the test SpeakerID.
3. Call Convai Delete Speaker ID with that value.
4. Clear the local saved SpeakerID.

Deleting a Speaker ID removes the user record and associated long-term memory records for that user across characters. Use this for confirmed test data or user-requested deletion flows.

Expected outcome: The test SpeakerID is removed from Convai and the local saved identity is cleared.

Next steps

End-user identitySpeaker ID managementTroubleshoot long-term memoryLTM Blueprint reference