How to read this list
A subprocessor processes personal data on ML BOOST’s instructions when ML BOOST is itself processing it for a business customer. A service provider may instead process data for ML BOOST as controller, or act as an independent controller for part of its service, such as payment compliance. The contract and feature determine the exact role.
This public list describes the current code-supported hosted path. It is not a promise that every listed feature is enabled in production or that every provider receives a particular customer’s data. A business order and the incorporated DPA should record the actual providers and region approved for that workflow.
Core hosted providers
| Provider | Purpose | Data involved | Route and location note |
|---|---|---|---|
| Google Cloud and Firebase Site host and French contracting entity: Google Cloud France |
Website hosting, Cloud Run compute and instance-local temporary storage, Firestore, Firebase Authentication, network delivery, and operational logs. | Public enquiry records; account and authentication identifiers; hosted session files processed on the Cloud Run instance; service metadata and logs. | Core for the current hosted service. Session files currently use access-controlled instance-local storage rather than a durable customer document bucket. Contracting entity, service location, support access, and transfers can depend on the account and Google service. No universal Pagewrite residency claim is made. |
| Stripe The applicable Stripe group entity is identified in the hosted checkout terms. |
Checkout, payment processing, fraud prevention, refunds, and payment records. | Account/order identifiers, product, amount, currency, billing and payment information entered at Stripe. ML BOOST does not receive the full card number. | Used for a paid hosted route. Stripe may act as a processor and as an independent controller for its legal, fraud, and payment-network duties. |
| Neon managed PostgreSQL | Managed relational database for deployments configured to use it. | Account, billing, entitlement, service-session, and technical-receipt records stored in PostgreSQL. | Used by the current hosted backend configuration. The selected database region and support path must be confirmed for a contractual residency requirement. |
Feature-dependent and optional providers
| Provider | When used | Data involved | Boundary |
|---|---|---|---|
| Mistral AI service | Hosted guided dialogue and, depending on workflow, OCR, vision mapping, metadata extraction, or advisory final review. | A blank or uploaded form, page image, extracted text, chat content, typed facts that enter free text, or bounded review content required for the selected operation. | The current consumer guided engine uses hosted Mistral. Final review masks field values by default, but free-text answers may still be present in dialogue history. Do not assume local processing. |
| Google Places Platform | Only when optional address suggestions are enabled and the user enters an address query. | The partial address query and technical request data needed to return suggestions. | Address suggestions are optional. The selected address can be entered without treating the suggestion provider as the authority for its accuracy. |
| PostHog service | Server-side product analytics only when a deployment has been configured with a PostHog API key. | Product event, pseudonymous account or session identifier, product/form identifiers, coarse counts, payment status, stage, and error code. Raw email addresses, notification contacts, form answers, and file contents must not be sent. | Disabled by default in code when no key is configured. The current integration does not require a browser-side PostHog script or cookie. The configured default host is the EU ingestion endpoint. |
| Telegram service | Restricted operator alerts when enabled. | Lead source and stable record link; for other operational events, an account identifier and technical or payment metadata may be included. Form answers and file contents must not be included. | The public lead alert deliberately omits contact and workflow content. Telegram is not a customer document-delivery channel unless separately agreed and assessed. |
Retention, model use, and provider terms
Pagewrite’s current application-level retention periods are stated in the privacy notice. Providers may keep security logs, backups, payment records, or service inputs under their own contract and account configuration. Their period may not match Pagewrite’s active-record expiry exactly.
No unsupported zero-retention or no-training claim. This page does not promise that a hosted model, OCR, or support provider has zero retention or never uses service data. The applicable provider terms and Pagewrite account settings must be reviewed for the intended workflow. A stricter promise must appear in the business order after it has been verified.
Customer-run and on-premises deployments
Pagewrite can be run in a customer-controlled environment. Local options such as Tesseract for OCR and Ollama for a model can remove some hosted providers from the Pagewrite processing path, with possible quality differences. The customer chooses and administers its infrastructure, identity, storage, logs, backup, model, OCR, and deletion configuration.
A locally deployed Pagewrite service can still send data out through the customer’s agent, model endpoint, OCR provider, observability stack, support request, or other connected system. “On-premises” describes the verified complete deployment, not merely where the Pagewrite process runs.
Provider changes and questions
We update this page before or when a material new hosted provider is added to the published data path. Notice and objection rights for a business customer are those agreed in its DPA or order. Removing a provider or changing a provider’s name may be reflected without advance notice.
To confirm the provider and region planned for a specific workflow, contact contact@ml-boost.com before sending live data.