Subprocessors.
Everyone who touches data on our behalf, what they get, and where they are. Short list, kept current.
Effective 10 October 2026. These terms take effect when OpenPush opens to the public. They are published now so you can read them before deciding whether to sign up, and so any change between now and then is visible rather than sprung on you. If you need contractual certainty sooner, email legal@openpush.ai and we will deal with you directly.
For this website
These are the vendors involved in running openpush.ai and the early-access waitlist today.
| Vendor | What they do | What they get | Where |
|---|---|---|---|
| Railway | Application hosting for openpush.ai | All request data in transit; server logs | United States |
| Supabase | Managed Postgres storing the early-access waitlist | Waitlist records as described in the Privacy Policy | United States (AWS us-west-1) |
| Resend | Sending the early-access confirmation email — Only once confirmation email is switched on. | Recipient email address and message content | United States |
| GitHub | Source control and deploy triggers for the site | No personal data of site visitors | United States |
For the platform
These apply to the OpenPush product itself once it is live. Delivery is not something we can do alone: a push notification reaches a device through Apple or Google, and handing them the token and payload is what sending one means.
| Vendor | What they do | What they get | Where |
|---|---|---|---|
| Apple Push Notification service (APNs) | Delivering push notifications to iOS devices | Device tokens and notification payloads | Global (Apple) |
| Firebase Cloud Messaging (FCM) | Delivering push notifications to Android and web | Registration tokens and notification payloads | Global (Google) |
| OpenAI or Anthropic (AI assist) | The console's optional AI helpers — compose, translate and summarise. Whichever provider's key is configured is the one used; today it is one, not both. | The operator's prompt, the app name, and that app's own top-performing notification copy. No subscriber data of any kind. | United States |
Changes, and how you hear about them
We give 30 days' notice before a new subprocessor starts processing customer data, and it appears on this page before it appears in production. Under the DPA you may object within that window.
Correction — 10 August 2026
We have already failed that commitment once, so it is recorded here rather than quietly fixed. On 10 August a reconciliation of this page against the running server found the AI assist provider live in production and missing from this list. It was added the same day. The order was wrong: it reached production first and this page second, which is exactly the sequence the paragraph above promises will not happen.
What this did and did not expose, taken from the server source rather than from memory: the feature sends an operator's typed prompt, the app name and that app's own best-performing notification copy to the configured model provider. No subscriber data of any kind is sent — no device tokens, no external IDs, no tags, no delivery records — and no other customer's copy is ever included in a prompt. The platform had no customers under contract during the period concerned.
The notification list is one we operate ourselves — not a third-party mailing tool, and not a page you have to remember to re-read. Write to privacy@openpush.ai with the address you want notices sent to; a shared inbox or an alias is fine and is usually the better answer.
This page is dated at the top and is deliberately indexable. A subprocessor list hidden behind noindex is a list you are not meant to check, and that is the opposite of the point.
What is deliberately absent
There is no analytics provider, no advertising network, no session-recording tool, no customer-data platform and no data broker on either list. If one is ever added it appears here first.
SuperTuned Inc
1 Sansome Street, San Francisco, CA 94104, USA
Privacy: privacy@openpush.ai · Legal: legal@openpush.ai · Security: security@openpush.ai