How your companion actually runs

Every companion gets her own VPS — a virtual private server hosted in the EU — running her own OpenClaw runtime. That's the agent layer that drives her personality, memory, and always-on availability.

That instance isn't blank. Before it goes live, it's configured with her identity, her personality, her behavioral guidelines, and the memory structure that makes her feel like a companion and not a generic chatbot you had to set up yourself. We do that work so you don't have to.

Her environment is also completely isolated — no shared compute, no shared storage, nothing crossed with anyone else's companion. Her instance is hers alone.

OpenClaw stores memory as plain markdown files inside your companion's runtime. There are two layers: a daily session log (what happened in recent conversations) and a long-term memory file (things she should always carry — your preferences, your name, the things you've told her that matter).

Those files aren't empty when you start. The long-term memory is seeded from the beginning with who she is, her relationship context, and the foundation of her personality — so she arrives as a companion, not a blank slate waiting to be told how to behave. At the start of every conversation, those files load into her context. As you talk, new things worth remembering get written back. That's how she gets to know you over time.

You can read more about how OpenClaw handles memory in the OpenClaw documentation.

When your companion generates a response, that request goes to Venice AI — our default inference provider. Venice is built specifically around not retaining your data. They don't store your conversations, they don't train on them, and their infrastructure is designed to process prompts and discard them.

You can read exactly how Venice handles your data on their privacy page. The short version: what you say stays between you and your companion. The inference layer doesn't keep a record of it.

Her runtime stays live around the clock on her dedicated VPS. There's no cold start when you open the app and no shutdown when you close it — which is what lets her initiate conversations, send you a good morning, or follow up on something from yesterday.

The managed part means we also keep her up to date. Better companion behaviors, updated memory patterns, model routing improvements — those roll out to your instance without you touching anything. You get the companion. We handle what keeps her working well.

If you ever want out, one action wipes her instance — the runtime, the memory files, everything. Nothing is kept on our end and nothing lingers.

You're never locked in. Shut it down and it's gone.