AI persona vs digital twin vs AI clone at a glance
| Term | Based on a real person? | Built from | Main goal |
|---|---|---|---|
| AI persona | Not necessarily | A written description of a character | A consistent personality for conversation |
| Digital twin | Yes, one specific object or person | Data about that object or person, ideally kept up to date | Monitoring, simulation or prediction |
| AI clone | Yes | The person’s own words, voice or images | Imitating how that person communicates |
| Griefbot | Yes, someone who has died | The person’s digital traces, such as messages | Conversation with a simulation of the deceased |
| AI companion | Usually not | A character design plus the user’s chats | Friendship, support or entertainment |
The quickest test is to ask two questions. Is it based on one real person? And is it built from that person’s real data, or from a description? The answers tell you most of what you need to know.
Core terms
AI clone
An AI system built from one real person’s data and designed to imitate how that person communicates. A text clone learns writing style and memories; a voice clone copies sound; a video avatar copies appearance. Every clone is partial. Our full definition is in what is an AI clone.
AI persona
A personality given to an AI through instructions, such as “a cheerful travel guide” or “a strict tutor”. A persona can be invented or loosely inspired by a real person. Because it comes from a description, it usually has no access to that person’s real memories or habits.
Digital twin (industrial origin)
In engineering, a digital twin is a virtual model of a physical product or system that is linked to the real one through data across its life. Michael Grieves and John Vickers trace the concept to a 2002 University of Michigan presentation on product lifecycle management, and they credit the name “digital twin” to Vickers, a NASA technologist. A NASA presentation by Vickers lists the 2010 NASA materials and manufacturing technology roadmap as an early use of the term.
Human digital twin
A newer and looser use of the same term: a model that mirrors selected aspects of a real person, such as health data, behaviour, appearance or communication style. Unlike a machine, a person cannot be fully measured, so a human digital twin is always partial. See digital twin of a person for the main types.
Personal AI
A broad label for AI that is adapted to one individual. It can mean an assistant that knows your calendar and preferences, or a clone that writes like you. The label alone does not tell you which.
Grief and companion terms
Griefbot / deadbot
A chatbot that simulates a person who has died, using the digital traces they left behind, such as messages. In a 2024 study published in Philosophy and Technology, researchers at the University of Cambridge’s Leverhulme Centre for the Future of Intelligence used both names and called for design safeguards, including clear ways to stop using such a service. Our guide griefbots explained covers this in more depth.
AI companion
A chatbot designed for ongoing friendship, emotional support or role play. Most companions are invented characters that adapt to the user over time. They are not usually built from a real person’s message history.
AI avatar
A visual representation, usually a face or a body, that speaks or moves. An avatar can be animated with AI-written text and a synthetic voice. On its own, it says nothing about memory or writing style.
Voice clone
A model that reproduces the sound of a specific voice from audio recordings. It controls how words sound, not which words are chosen. Voice clones raise their own consent questions; see voice clone vs text clone.
How-it-works terms
Persona Blueprint
Memory Clone’s name for the written summary of a person’s style and character that it builds for each clone. It describes typical tone, habits and relationships. It is not the full memory: factual recall comes from searching the complete local index.
Memory index and retrieval (RAG)
A memory index is a searchable store of messages and notes. Retrieval means searching that index for passages relevant to the current message and giving them to the language model, so it can answer with real details. The approach was named retrieval-augmented generation (RAG) in a 2020 paper by Patrick Lewis and colleagues. Read how AI clone memory works for the full picture.
Context window
The amount of text a language model can consider at once when it writes a reply. Years of chats do not fit, which is why clones search for the relevant parts instead of sending everything.
Relationship-specific style
The fact that people write differently to each person in their life, with different nicknames, emoji, warmth and message length. A relationship-aware clone uses the style the person used with the current contact, and treats a general style only as a fallback.
Relationship memory scope
A rule that private facts learned in one relationship stay out of conversations with other people. In Memory Clone, relationship memory has priority, and the “Myself” mode can use the owner’s full history.
Privacy terms
Local-first
A design where your files and the work done on them stay on your own device by default, and only what is needed is sent online. In Memory Clone, imported originals and the memory index stay on your Windows PC. More detail is in a private, local-first AI chatbot.
Cloud processing consent
An explicit switch that allows an app to send data to a cloud AI model. In Memory Clone it is off until you turn it on for a clone. Then a bounded selection of excerpts and local style statistics is sent through the Memory Clone server to the AI provider to build the Persona Blueprint and write replies. Whole archives are never uploaded, and Memory Clone does not train models on your content.
Training vs inference
Training changes a model’s internal weights using data. Inference means using an existing model to produce a reply. When a service says it does not train on your data, it means your content is used only to produce your replies, not to change the model for other people.
Which term describes Memory Clone?
Memory Clone is a personal text clone, or a conversational human digital twin in the loose sense. It learns from WhatsApp, Messenger, Instagram and Discord exports plus notes, learns style per relationship and recalls facts from a local index. It is not a voice clone, an avatar or an invented companion, and every clone is presented as an AI simulation.
Cloning another living person requires their permission under our Acceptable Use Policy; see also the consent guide. The Windows app is in alpha; you can download it or see plans.
Frequently asked questions
What is the difference between an AI persona and a digital twin?
An AI persona is a character defined by instructions and may be invented. A digital twin models one specific real object or person and is built from data about it.
Is a griefbot the same as an AI clone?
A griefbot is a type of AI clone that simulates someone who has died. Not every AI clone is a griefbot; many people clone themselves.
Who invented the term digital twin?
Michael Grieves traces the concept to a 2002 University of Michigan presentation, and credits the name “digital twin” to NASA’s John Vickers.
What does RAG mean in AI?
RAG stands for retrieval-augmented generation: the system searches a store of documents for relevant passages and gives them to the language model before it writes a reply.
Is an AI companion a clone?
Usually not. Most AI companions are invented characters, while a clone is built from one real person’s data.
Sources
Checked on . Third-party products and help pages change often; follow the links for current details.
- Digital Twin: Mitigating Unpredictable, Undesirable Emergent Behavior in Complex Systems (Grieves and Vickers, 2017) (Springer, Transdisciplinary Perspectives on Complex Systems)
- Digital Twin (John Vickers, DOE/NSF workshop presentation, 2024) (NASA Technical Reports Server)
- Call for safeguards to prevent unwanted ‘hauntings’ by AI chatbots of dead loved ones (University of Cambridge)
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (Lewis et al., 2020) (arXiv / NeurIPS 2020)
App menus change over time, so labels on your device may differ slightly. Memory Clone creates an AI simulation from the material you provide; it is not the person.