Subprocessors

About this list

ERP•AI US Inc. (“ERP•AI”, “we”, “us”), 2 Embarcadero Center, San Francisco, CA 94111, USA, provides the ERP•AI platform, Proto, build.host, APIs, websites and related support (the “Services”). Capitalized terms not defined here have the meanings in our Terms of Service and our Data Processing Addendum (the “DPA”).

A Subprocessor is a third party, including an affiliate, that we engage to process Customer Personal Data (as the DPA defines it, including personal data in Customer Content and Hosted Content, logs your hosted apps produce and sign-in data of people using your Published Apps).

This page is the Subprocessor list that the DPA refers to. By agreeing to the DPA, you authorize the Subprocessors listed here on that date, and later changes made as described in “Changes and notice” below. The DPA governs our obligations and prevails if this page conflicts with it. Personal data that ERP•AI controls, such as account, billing, support and website data, is covered by our Privacy Policy; some providers below also process it, as the Role column says.

Reading the tables

• Role. “Subprocessor”: processes Customer Personal Data for us; the DPA sets the terms we must impose on it. “OpenRouter’s subprocessor”: a model host engaged by OpenRouter, not ERP•AI. “Own terms”: receives data under its own terms, not as our Subprocessor. “Service provider”: processes only data ERP•AI controls, for us. “Third party”: uses data for its own purposes.
• Services. Platform (apps.erp.ai, its APIs and admin console), Proto (the app and the ERP•AI servers it uses), build.host (including hosted sites) and Websites (www.erp.ai, erpai.studio, protoagent.ai, proto.erp.ai and status.erp.ai). “All” means all four.
• Location. “Global” means the provider’s worldwide network. “Not restricted” means we have not limited where the provider processes the data.
• A feature’s provider receives data only when the feature runs for your account, including through an Agent or by default (such as Proto Managed and Proto’s web search).

At a glance

• Our main servers, including our self-hosted platform databases, are Hetzner servers in Ashburn, Virginia and Hillsboro, Oregon, USA.
• Uploaded files and backups in Cloudflare R2 may be stored outside the United States, and record exports are stored with Amazon Web Services in Singapore.
• AI features send prompts and the content a task needs to Model Providers, mostly through OpenRouter. Model hosts may process requests outside the United States, and some we route to, including Z.AI and Moonshot AI, are headquartered in China.
• We do not use Customer Content to train or fine-tune AI models.

Infrastructure and core services

Sites hosted on build.host are served from our Hetzner servers, not through Cloudflare (though some pages uploaded from Proto load a script through it), and Cloudflare does not carry traffic between the Proto app and our Proto servers. Microsoft Azure Container Registry, Let’s Encrypt (TLS certificates), Open Exchange Rates, Namecheap and Apple and Microsoft code signing receive no Customer Personal Data and are not Subprocessors.

AI model providers

Model Providers receive the prompts and Customer Content an AI feature needs, and return Output.

How we route model requests

• Provider settings. For in-app Agent requests through OpenRouter, we use OpenRouter’s setting that excludes model hosts that, as OpenRouter classifies them, store request data or use it for training; some may still keep data briefly, for example to monitor abuse. Other requests we route on our account, including Proto Managed and other platform AI features, do not use that setting today, so OpenRouter may send them to hosts that keep request data or use it to improve their models. Requests we send directly to Anthropic, OpenAI or another Model Provider are handled under that provider’s terms with us.
• Proto Managed is Proto’s default model mode; we set its model on our servers and may change it. Our Proto servers forward to OpenRouter or Fireworks: Proto Managed requests; requests made with your own OpenRouter or Fireworks key; planning, decision and workflow-writing requests in every model mode, including Codex; and page-building, schema and image-generation requests.
• What our Proto servers log. These requests carry your user ID, email address, organization, app and session identifiers. Our servers log each one: the request body in full up to 16 KiB (a summary above that), the response up to 64 KiB, those identifiers, the model, token counts and cost. They use your provider key only to call that provider and are not designed to store it. We keep these logs as long as we need them to bill, support you, investigate problems and secure the Services, or as the law requires.

Model Provider usage policies

When we route your requests to a Model Provider, its usage policy also applies, as our Acceptable Use Policy explains. Where a provider has no separate usage policy, we link its terms.

• OpenRouter: Terms of Service, which also apply the terms of each host it lists
• Atlas Cloud: Acceptable Use Policy
• Weights & Biases: Terms of Service; CoreWeave: Acceptable Use Policy
• Parasail: Terms of Service
• DeepInfra: Terms of Service
• Cerebras: Terms of Use
• Baseten: Terms and Conditions
• Z.AI: Terms of Service
• Moonshot AI: Model Use Agreement
• Anthropic: Usage Policy
• OpenAI: Usage Policies
• Fireworks AI: Terms of Service

Other feature providers

Providers for our websites and operations

These providers process personal data that ERP•AI controls, mostly about website visitors, not Customer Personal Data. Advance notice does not apply to them, but we update this page when they change. Our Cookie Policy explains the cookies and choices involved.

protoagent.ai runs no analytics. Unsplash (website images), YouTube, Vimeo and Loom (videos users add to a knowledge base) and jsDelivr (a library in downloaded plan files) receive your IP address when your browser loads their content. Our Google Tag Manager container also holds a LinkedIn Insight Tag, an RB2B visitor-identification tag and a second PostHog tag, which our sites do not trigger and whose scripts our security policy does not allow; before any of them runs, we will list it here and in the Cookie Policy.

Where ERP•AI’s own systems run

These systems run on the Hetzner servers listed above.

Access by our personnel. Our personnel, and those of companies that help us run the Services, may access Customer Personal Data from the countries where they work, including outside the United States, to support and operate the Services. The DPA governs that access.

Proto content that reaches our servers. Proto keeps your files, session history and memory on your computer, but when an agent error occurs it sends our Proto servers an automatic error report containing the full session transcript (every message and session entry, the session title and your workspace folder path, with image data removed). There is no in-app setting to turn this off; in modes without a connection to our servers, reports wait on your computer until there is one. Rating a reply down sends that message and the session log. Problem reports include your chat unless you clear the “include chat” box; we pass them, without the chat, to a support app on the platform. Proto also sends short error events, with credentials removed, in every model mode. The Proto Terms describe these flows.

Third-Party Services you choose

When you connect a Third-Party Service, we send data to it on your instructions, and it processes that data under its own terms with you, not as our Subprocessor. For example:

• Your own model keys in Proto (OpenAI, Anthropic, Mistral, Azure OpenAI, Azure AI Foundry, Google Gemini, Cloudflare Workers AI, OpenRouter, Fireworks or a custom endpoint). Most requests go directly from your device to the provider. OpenRouter and Fireworks key requests pass through our Proto servers and are logged as described above, and the Proto services above may use our servers whatever key you use.
• Codex with a ChatGPT account. Proto sends its main model requests directly from your device to OpenAI, and your ChatGPT tokens only to OpenAI, but still uses our servers for the Proto services and error reports described above. Each time Proto starts, in any model mode, if it has no ChatGPT sign-in of its own, it copies any Codex CLI sign-in tokens on your computer into its own settings file and may refresh a copied token. OpenAI’s terms for your ChatGPT plan, and on individual plans your ChatGPT data controls, govern OpenAI’s use of your content.
• Platform connections you set up: your own OpenRouter key, your own S3 bucket (reached with temporary AWS credentials) and your company’s single sign-on provider.
• Workflow and app integrations that use your credentials, such as Mailgun, SendGrid, Twilio, Microsoft 365, Salesforce, Google APIs, Slack and IMAP mailboxes; some use ERP•AI’s own Microsoft and Salesforce app registrations.
• Proto plugins and connections, where available: Google Workspace (its sign-in tokens stay on your device), messaging bridges and MCP servers.
• Websites and apps an Agent visits or acts on for you.

How we oversee Subprocessors

• Terms and responsibility. The DPA sets out the data protection terms we must impose on each Subprocessor and our responsibility to you for its performance of them.
• International transfers. For Customer Personal Data from the European Economic Area, the United Kingdom or Switzerland, the DPA applies the EU Standard Contractual Clauses, the UK International Data Transfer Addendum and the Swiss amendments, and governs onward transfers to Subprocessors. ERP•AI is not certified under the EU-US Data Privacy Framework.
• US state laws. This page and our notices under “Changes and notice” are our notice to you of each Subprocessor we engage to help process personal information for you (Cal. Civ. Code 1798.140(ag)(2)). Where another US state law lets you object to a processor’s subcontractors, use the objection process below.

Changes and notice

New Subprocessors. At least 30 days before a new Subprocessor starts processing Customer Personal Data, or a listed Subprocessor starts processing it in a new country, for a new purpose or Service or in new categories of Customer Personal Data, we will update this page, add the change to the change log and email everyone subscribed to updates. The DPA describes any other notice we give.

Subscribe. Anyone can receive these notices: email legal@erp.ai with the subject “Subscribe to subprocessor updates” and we will add you to the notice list.

Objections. You may object on reasonable data protection grounds by emailing legal@erp.ai within 30 days of our notice, and we will work with you in good faith on a solution, such as avoiding the feature that uses the Subprocessor. While a timely objection is open, we will not use the new Subprocessor for your Customer Personal Data where that is technically feasible; where it is not, we will tell you before it starts and you may end the affected Services. If we cannot resolve your objection within 30 days of receiving it, you may end the affected Services, or close your account, by notice to legal@erp.ai, and we will refund unused Purchased Credits and prepaid fees for the ended Services as the DPA describes.

Urgent replacements. If a Subprocessor fails, stops its service or creates a security risk, we may replace it sooner, where possible with a listed provider, and otherwise with one doing the same job in the same country or under the same transfer safeguards. We will notify you as soon as we can, before the replacement starts where possible and within 10 business days after it starts at the latest, and the objection rights above apply from that notice. Where the EU Standard Contractual Clauses apply to your data, they prevail over this paragraph.

Removals and corrections. When we stop using a provider, or correct a clerical error that does not change what a provider does, we update this page and the change log without advance notice.

Order Forms. If your Order Form or signed DPA sets different notice terms for Subprocessor changes, those terms apply to you.

Change log

Contact

Send questions, objections and subscription requests to legal@erp.ai, or by post to ERP•AI US Inc., 2 Embarcadero Center, San Francisco, CA 94111, USA. Report security issues to legal@erp.ai.