Subprocessors

Draft for owner and legal review. The routing, versioning, and acceptance mechanics around this document are production code. The prose is not legal advice and has not been reviewed by counsel.

These are the companies that process data on PpcGPT's behalf. We publish a dated entry here and email account owners 30 days before adding one.

Provider	Purpose	Data category	Region
Vercel	Application hosting	Request metadata	United States
Supabase (on AWS)	Authentication, database, file storage	Account data, Amazon business data, uploaded invoices	United States
Trigger.dev	Scheduled and background job orchestration	Job metadata, store identifiers	United States
Anthropic	Language-model processing for drafting and document reading	Business metrics and document text, with no tenant identifiers attached	United States
Resend	Transactional email	Email address, message content	United States
Better Stack	Uptime monitoring	Service availability only	United States
GitHub	Source control and continuous integration	No customer data	United States

Anthropic does not train on PpcGPT customer data, and the prompts PpcGPT sends carry no account, store, or seller identifiers.

Corrections

2026-07-31 — Sentry removed. An earlier version of this page listed Sentry as an error-monitoring subprocessor. It never was one: no Sentry integration was ever deployed and no data was ever sent to it. Error monitoring is not currently provided by a third party. When one is added it will appear in the table above, with 30 days' notice as described at the top of this page.

Not currently subprocessors

Railway is not on this list. It appears in our architecture notes for a keyword-research experiment that is disabled and not provisioned. If that experiment is ever activated we will add Railway here first, with the purpose "experimental keyword worker" — and it would receive no Amazon data and no tenant credentials.

Contact

Questions: privacy@ppcgpt.ai.