AI companion vs AI assistant vs agent vs chatbot
A chatbot is a conversational interface, an AI assistant answers requests on demand, an AI agent carries out multi-step tasks toward a goal, and an AI companion is a persistent presence that remembers you across sessions and devices. The words overlap in marketing, but the designs behind them differ in session model, memory, and initiative, and the right choice depends on whether you need answers, execution, or continuity.
Four words that get mixed together
Marketing uses chatbot, assistant, agent, and companion almost interchangeably, which makes real comparisons hard. The four words actually answer four different questions. Chatbot answers: what is the interface? Assistant answers: what is the service model? Agent answers: how does the work get done? Companion answers: what kind of relationship does the product hold with you over time? Once you separate the questions, most product claims become easy to classify.
Precise definitions
- Chatbot: a conversational interface. You type or speak, it replies in turn. The word says nothing about memory, capability, or autonomy; it names the surface, not the system.
- AI assistant: a system that answers requests on demand. You ask, it responds, the exchange ends. Voice assistants on phones and smart speakers are the spoken form: command in, action or answer out.
- AI agent: a system that pursues a goal through multiple steps, choosing actions along the way, such as searching, calling tools, or writing files, usually with checkpoints where a human approves.
- AI companion: a system designed for continuity. It keeps durable memory of you, is present across your devices, and can raise things at the right moment rather than only when asked.
The comparison at a glance
| Kind | Session model | Memory | Initiative | Privacy stakes |
|---|---|---|---|---|
| Chatbot | One conversation at a time | Usually within the conversation only | None; it waits for input | Low if nothing persists, higher if transcripts are retained |
| AI assistant | Request and response | Little to none between requests | None; it acts when asked | Moderate: requests reveal intent even when nothing is stored |
| AI agent | A task, start to finish | Working state for the task | High within the task, none outside it | Depends on what systems it can touch while acting |
| AI companion | An ongoing relationship | Durable, across sessions and devices | Measured: surfaces relevant things from your own data | Highest: it accumulates a long-lived record of your life |
Where the lines blur
These categories compose rather than compete. A companion almost always contains an assistant: you can still just ask it things. A companion may use agent-style execution for a multi-step job you hand it. And nearly all of them wear a chatbot interface, because conversation is the natural surface. The classification that matters is the deepest layer a product actually implements. A chat window with no memory is a chatbot no matter what the landing page calls it, and a product that quietly keeps a durable profile of you is a companion in architecture even if it is marketed as a simple assistant. Judge products by the layer, not the label.
Which to choose when
| You mainly need | Choose | Why |
|---|---|---|
| One-off answers, drafts, and explanations | An assistant behind a chatbot interface | Statelessness is a feature here: nothing accumulates, nothing needs protecting |
| A defined multi-step job done for you | An agent, with approval checkpoints | Autonomy is what completes tasks, and checkpoints keep it accountable |
| Help that follows your actual life over time | A companion | Continuity is the value: calendar, tasks, reminders, and memory in one presence |
| Continuity, but you cannot accept a readable server copy of your life | A companion with end-to-end encrypted storage | Persistence is the risk, so the stored record itself must be sealed on your device |
Why the distinction matters for privacy
Privacy stakes scale with persistence. An assistant that forgets each request can leak only that request. A companion that works well accumulates your schedule, your health notes, your money, and your worries, and keeps them for years. That dataset is the product's value and its risk at the same time, which is why the storage architecture question, who can read the stored record, matters far more for a companion than for any other kind. It is also why initiative deserves scrutiny: a companion should derive its proactive moments from data you already handed it, such as your own calendar, not from monitoring channels you never offered.
How Violet fits
Violet is built as a companion, currently pre-launch. The continuity layer is there: durable memory, a daily view built from your real calendar, tasks, and reminders, and proactive lines computed only from that data, with an honest quiet line when the day is empty. The privacy answer to the persistence risk is architectural: stored content is sealed on the device with a key derived from your passphrase, and the sync server holds only envelopes of ciphertext it cannot open. Voice is push-to-talk: the microphone is live only while the button is held, with no wake word and no always-listening mode. A Trust panel in the app lists what is actually stored, read from real state.
Common mistakes
- Calling any chat interface a companion. Without durable memory and cross-device continuity, it is a chatbot, whatever the branding says.
- Treating agent as simply better than assistant. Autonomy is a tool for tasks, not an upgrade tier; for a quick answer, an agent is overhead.
- Assuming memory requires the provider to read your data. Durable memory can be stored as ciphertext and decrypted only on your devices.
- Assuming proactivity requires surveillance. A companion can compute its nudges from data you explicitly gave it, and stay silent otherwise.
- Choosing by demo instead of by session model. A demo shows one exchange; the categories differ in what happens between exchanges.
Questions
Is a voice assistant on a phone or smart speaker an AI companion?
By these definitions, no. Built-in voice assistants are spoken command assistants: a request comes in, an action or answer comes out, and the exchange ends. A companion is defined by durable memory and continuity across sessions and devices. The interface being voice does not change the category; the session model does.
Can one product be an assistant, an agent, and a companion at once?
Yes, and good ones are layered exactly that way. A companion contains an assistant, because you can still just ask it things, and it may run agent-style multi-step execution for jobs you hand it. The label that matters is the deepest layer the product actually implements, because that layer determines both its value and its privacy stakes.
Which type collects the most sensitive data?
The companion, by design. Its value comes from accumulating a durable record of your life, which is exactly the data most worth protecting. That is why storage architecture is the first question to ask of any companion: whether the stored record is readable by the provider, or sealed on your device with end-to-end encryption.