It's the first question almost every customer asks, and it's the right one. Across the industry, "we may use your data to improve our models" usually means your conversations are collected into a dataset, shipped to a lab, and used to train the next model that everyone else uses too. Opting out, if it's possible at all, is buried in settings.
Voltaic is different, and because it genuinely learns from use, we owe you a precise answer rather than a reassuring one.
The short answer
Voltaic never harvests your conversations into a dataset to train a shared model. The spiking core adapts its own weights locally, inside your workspace, and only from what you deliberately teach it.
Why the question is different for Voltaic
Most AI models are frozen at training time. They can't learn from you, so "training" can only mean collecting your data and building a new model somewhere else. Voltaic learns continually, at inference time. That's its whole point. So yes, it does learn, but where that learning happens, who it benefits and what it can touch are completely different.
What actually happens
- You deliberately teach a fact or share a document. Learning starts with an intentional act, not with background collection.
- The spiking core adapts its own weights, in place, inside your workspace. No copy is sent anywhere to be learned from.
- What it learns is written into your workspace's private memory. You can search it, pin it, edit it and delete it.
- Recall improves, for you only. Other customers' models don't get better from your data, and yours doesn't absorb theirs.
What doesn't happen
- Your chats are not silently swept into a training set.
- Your data is not shipped off to a distant lab.
- It is not baked into a foundation model that serves other customers.
- It is not sold or used to train a competitor's model.
Side by side
- When: conventional training happens offline, in a distant data centre. Voltaic learns live, as you use it, scoped to your workspace.
- Where your data lives: conventionally, pooled into a shared dataset. With Voltaic, it stays in your workspace, unless you choose to allow cloud routing.
- What learns: conventionally, a single model shipped to everyone. With Voltaic, only your workspace's memory and weights.
- Who benefits: conventionally, the vendor's next model. With Voltaic, you, and only you.
- Opting out: conventionally, buried in settings if it exists at all. With Voltaic, there's nothing to opt out of.
No replay buffer, by design
Many continual-learning systems avoid forgetting by keeping copies of old data and re-training on them. That would mean storing your data indefinitely just to stay accurate. Voltaic's benchmark results are achieved with no replay buffer: Hierarchical Synaptic Consolidation and HULL protect old knowledge inside the weights themselves, so there's no hidden archive of your history to manage.
Controls you keep
Privacy as a default is the starting point, not the whole story. Verkko also gives you full export, deletion and data residency controls, including EU deployments. You can run in Verkko Cloud, in your own cloud, or fully self-hosted and air-gapped, on a single GPU. And when Voltaic does reach out to an external model for extra scale, your routing rules decide what can leave: pin sensitive topics to local-only, allow-list the cloud models you trust and set policies per workspace or per user.
The bottom line
Continual learning is scoped to your workspace. Your knowledge compounds for you, on your terms, and nowhere else. There's nothing to opt out of, because privacy is the default.