Made for Swiss self-employed & SMEsThe two models
| Local (self-hosted) | Hosted / cloud SaaS | |
|---|---|---|
| Data location | Stays on your hardware | Processed on a vendor's servers |
| Setup | You install & maintain it | Zero setup, vendor runs it |
| Cost | Free software; you pay model/API usage | Subscription + usage |
| Control | Full control, inspectable, extensible | Limited to vendor features |
| Privacy / DSG | Easier to keep client data close | Depends on vendor & data-processing terms |
| Best for | Sensitive data, tinkerers, long-term control | Speed, no maintenance, teams without IT |
Both OpenClaw and Hermes are local-first by design — you run the agent on your own machine or server. That's a natural fit for Swiss businesses that want client data to stay under their control.
Where your data actually goes
Under the Swiss Data Protection Act (DSG/nDSG), client data must be handled carefully — which is why a notary, doctor, fiduciary or anyone holding sensitive information should think about this before adopting an agent.
- The agent itself (the harness that reads your inbox and calls tools) runs on hardware you control.
- The model (the reasoning brain) usually runs in a vendor's cloud — unless you deliberately run a local model.
- Your memory and skills (what the agent has learned about you) live on your own machine.
So "local" can mean two things: a local agent (common, easy) or a local model (more work, but nothing sensitive leaves your premises).
What about the model itself?
Even a local agent usually calls a hosted model (like Claude or GPT) over an API. If you want the reasoning local too, you can run an open-weight model on your own hardware — Hermes supports local models natively, and OpenClaw has community adaptations for models like DeepSeek.