Subprocessors
Everyone who touches your data.
Last updated: 8 September 2026
A subprocessor is a vendor we engage that may process personal data in the course of running Composita. This is the complete current list. Each one is bound by a data processing agreement, and we keep dated copies of all of them.
We announce any addition or replacement on this page at least 30 days before it takes effect, and notify customers directly, so there is time to raise an objection first. This page is the reference for that commitment.
| Subprocessor | Purpose | Personal data | Location | Transfer mechanism |
|---|---|---|---|---|
| Hetzner Online GmbH | Cloud infrastructure — all platform data at rest | All categories | Germany (EEA) | Not applicable — data stays in the EEA |
| DigitalOcean, LLC | Crawler node, container registry, DNS | Alert metadata, images | Amsterdam (EEA); US parent | Data Privacy Framework incl. UK extension; Standard Contractual Clauses with UK Addendum as fallback |
| OpenRouter, Inc. | Model routing for drafting, verification, subtitle translation, art direction, and image generation | Editorial content submitted to models | United States | EU Standard Contractual Clauses (Module 2) with UK Addendum; zero-data-retention routing |
| AC PM LLC (Postmark) | Transactional email delivery of digests | Recipient email addresses, message metadata | United States | Data Privacy Framework; Standard Contractual Clauses (Modules 2–3) with UK Addendum |
| Google Ireland Ltd (Firebase Cloud Messaging) | Android push notification transport | Push payloads in transit (mentions, room names) | Global | Firebase Data Processing and Security Terms: Standard Contractual Clauses and Data Privacy Framework |
What is deliberately not on this list
Two parts of the platform that would normally involve a vendor do not, because we run them ourselves. This is a design choice, not an accident, and it is why the list above is as short as it is.
- Newsroom chat. The messaging server is self-hosted on our own EEA infrastructure. Messages, voice notes, and attachments do not pass through a third-party messaging platform.
- Monitoring inference.Relevance judging, enrichment, transcription, transcript cleanup, and embeddings all run on GPU hardware we own and operate. Content in the monitoring pipeline never leaves our infrastructure. Only editorial production — drafting, verification, subtitles, and images — reaches an outside model, through OpenRouter as listed above.
We also use a small number of vendors that never touch customer data — uptime checks, operator paging, and bookkeeping. They are tracked in our internal vendor register but are not subprocessors, because no personal data from the platform reaches them.
Questions or objections
If you want the detail behind any row, or you want to object to a planned change, write to privacy@composita.ai. How this fits together with the rest of our data handling is set out in our Privacy Policy.