Advanced
Living Memory: retention and identity matching
Version 1.0 · Last updated 2026-06-25
A customer who asked about pricing yesterday should not get the same first-time pitch today. A customer with a recurring issue should not have to re-explain it. Living Memory is what makes the agent feel like a continuous relationship instead of a stateless chatbot.
Key concepts
Identity matching
How the platform decides that two sessions belong to the same customer. Rules combine email, account ID, custom identifiers, and inferred identity from Living Context.
Retention windows
How long memory is kept. Different defaults for anonymous visitors (shorter) and identified customers (longer). Aligned to your sales cycle and GDPR data minimization.
Memory fields
What the agent remembers per customer: identity, conversation history, preferences, sentiment, goals, and the outcomes of past conversations. Each customer’s memory is private to that customer. Extensible with custom fields where needed.
Anonymous-to-identified linking
When an anonymous visitor identifies themselves (sign-up, login, form fill), the platform links their anonymous history to the identified profile.
Erasure
Customers can request memory erasure under GDPR. Erasure drops the customer’s history immediately; the next conversation starts fresh.
What you can do here
- Configure identity matching rules
- Set retention windows for anonymous and identified profiles
- Add custom memory fields
- Inspect what Living Memory holds for a specific customer
- Process data subject erasure requests
When to use it
- Default behavior is close but not quite right for your business
- Sales cycle is unusual (very short or very long)
- You need to remember something specific that defaults do not cover
- Privacy review wants tighter retention windows
How it works
Each Customer Agent conversation writes to Living Memory under the customer’s identity. Subsequent conversations read from Living Memory before generating answers. Memory passes through the Privacy Vault, so PII is tokenized.