Your data
Where your data lives.
What is stored, what leaves the platform, and what you can ask us for. If your case does not allow information to leave your network, Magnus is installed on your own infrastructure.
What is stored
Your account data (name, email, organization), the conversations your customers have with your agents, the documents you upload as a knowledge base, and the trace of every turn: what the agent understood, which rule it decided by, which tool it ran and what came back. The trace is what lets you answer “why did the agent do that” months later.
Who can see it
Every organization is isolated from the others: an agent, an API key and a user belong to one organization and cannot reach another one’s data. Inside your organization, the roles and permissions you configure are what rule.
What leaves the platform, and where to
The difference between these two columns is the reason the on-premise installation exists.
| Managed by us | On your infrastructure | |
|---|---|---|
| Conversations and traces | In our database, isolated per organization. | In your database. They never leave your network. |
| The text the model needs | Sent to the provider (OpenAI, with Google Gemini as fallback) to generate each reply, under their terms of service. The OpenAI API does not use that content to train its models. | With a local model — Ollama or any OpenAI-compatible server — it never leaves your network. |
| Your documents | Their text is sent to the embeddings provider so it can be searched. | With a local model, they do not leave either. |
| Voice: transcription and speech | Cloud providers (OpenAI and ElevenLabs). | Agreed at installation time. |
Where it runs
The managed version runs on dedicated Hetzner servers. Data handling follows Uruguay’s Law 18.331 and is aligned with the European GDPR and with United States data protection regulation. We hold no SOC 2 or ISO 27001 certification: if your purchase requires one, say so up front and we will tell you what we can evidence instead.
Magnus on your own infrastructure
If your case requires it — by regulation, by contract or by internal policy — Magnus is installed in full on your own servers. With a local model no conversation leaves your network: not the text, not the documents, not the traces. It is delivered as a project, with the scope agreed case by case, and the voice part is defined in that conversation.
Ask about an on-premise installationHow we improve the product
We review conversations to fix mistakes and improve the agents. You can ask us to exclude your organization’s conversations from that review.
What you can ask us for
Access, rectification, export or deletion of your organization’s data. The full privacy policy, with the statutory deadlines, is published separately (in Spanish).
Read the privacy policyUpdated October 22, 2026.