Subprocessors
Every external service that processes platform data, and what it sees.
| Service | Purpose | What it sees | Commissioned by |
|---|---|---|---|
| Cloudflare | Hosting and storage of the platform (compute, database, object storage, key-value store). | Ciphertext, the metadata listed in section 7 of the whitepaper, operational logs of the workers, and — in our worker's memory for the duration of a request — the cleartext of server-routed AI requests (whitepaper §6). | us — our subprocessor |
| Mailgun (EU region) | Transactional email — registration and recovery codes, invitations, account notices. | Recipient addresses and the mail contents named in section 7 of the whitepaper. | us — our subprocessor |
| Opper (AI gateway, Sweden) | Gateway for the platform AI models, run on AWS in Stockholm under a data processing agreement. It forwards a server-routed AI request to the processor of the platform model chosen for it. Reached only for a platform model your tenant administration has left enabled. | The cleartext of server-routed requests to a platform model — the single exception the whitepaper names. Request content is not stored: call tracing is switched off for our project, and only usage metadata (model, volume, cost) is kept for billing. | us — our subprocessor |
| Microsoft Azure (EU) | Runs the OpenAI platform model in the EU, reached through Opper. | The cleartext of requests to that model. Kept for up to 30 days for abuse monitoring; not used for training. | us — our subprocessor |
| Google Vertex AI (EU) | Runs the Anthropic platform model in the Netherlands and the Google platform model in the EU, both reached through Opper. | The cleartext of requests to those models. Anthropic model: may be kept for abuse monitoring, for a period the provider does not state. Google model: zero data retention. Neither is used for training. | us — our subprocessor |
| TensorX Ltd (Ireland) | Runs the open-weight DeepSeek platform model on hosting in Ireland and Finland, reached through Opper. Used only when a user selects this model explicitly; DeepSeek's own service is never called. | The cleartext of requests to that model. Zero data retention; not used for training. | us — our subprocessor |
| Mistral AI (France) | Runs the Mistral platform model on its own infrastructure in the EU, reached through Opper. | The cleartext of requests to that model. Zero data retention; not used for training. Content may be cached for replay, for a period the provider does not state. | us — our subprocessor |
| Your own AI endpoint | Processes AI requests to a model your tenant administration registers at an endpoint of its choice — for example your own model server, or a proxy in front of it. | The cleartext of each request sent to it; where the request is server-routed, that is the single exception the whitepaper names. | you — enters the chain only through your configuration |
| Your SIEM endpoint | Receives audit events if your tenant administration configures a webhook. | Audit metadata. Assessment content is not part of that stream. | you — enters the chain only through your configuration |
For the website itself (trausto.com), Cloudflare Pages and Resend apply; they are listed in the privacy policy.
Platform AI models
The models available on the server-routed AI path without an endpoint of your own. Each is reached through Opper and runs at the processor named in the table above; your tenant administration can disable any of them.
| Model | Vendor | Runs at | Note |
|---|---|---|---|
| GPT 6.1 Sol | OpenAI | Microsoft Azure (EU) | — |
| Claude Sonnet 5.5 | Anthropic | Google Vertex AI (EU) | — |
| Gemini 4 Argon | Google Vertex AI (EU) | — | |
| DeepSeek V4.1 Flash | DeepSeek | TensorX Ltd (Ireland) | only when selected explicitly |
| Mistral Large 4 | Mistral AI | Mistral AI (France) | — |
As of 2026-10-19. A new model version changes only this list; a new processor changes the table above.