Boundsmap

Your Replit app uses an AI model. Who gets your users’ data?

By Iain Barclay ·

Replit's documents as read on 21 October 2026. Platforms change their plans and terms, so check the current pages before you rely on any of this.

The short answer

It depends on your Replit plan. If your app calls an AI model through Replit AI Integrations, the call uses Replit's credentials and Replit bills you, and you don't need an account with the model's maker. So Replit's terms are the ones to read first.

Replit's data processing agreement (DPA) comes with its Pro and Enterprise plans and takes effect with the plan: there's nothing separate to sign. Core is covered by Replit's Terms of Service, which don't include a DPA. From our read, Replit's documents don't say whether the agreement covers the AI calls your app makes from any plan. Here we talk about what they do say, and what's still open.

Does your app use AI while it runs?

Replit Agent writing your code isn't your app using AI. If nothing in your app calls a model when someone uses it, this page doesn't apply to you.

If it does (a chat, a summary, a draft reply, a suggestion), and some of your users are in the EU or the UK, read on, wherever you are based. Boundsmap has rules for the EU and the UK, so that's the law this page covers.

Which plan are you on?

Here is what Replit's pages say for each plan.

Replit's data processing agreement (dated 1 April 2026) says it "forms part of and is incorporated into the Replit Commercial Agreement" and "will take effect on the effective date of the Agreement". The Terms of Service (dated 3 August 2026) say that agreement applies "If you use our Pro or Enterprise Service".

How to tell which route your app uses: Agent's first response shows a label such as "OpenAI (Replit managed)" when it is using Replit's credentials, and an API key in your Secrets means it is using your own. Replit's AI Integrations page lists "(Replit managed)" labels for OpenAI, Anthropic, Google and OpenRouter.

Who receives your users' data?

There are three cases, and under the GDPR, in the UK and the EU, each company's role follows what it actually does with the data, not the paperwork.

With your own API key, on any plan. Your agreement is with the model's maker. An account your company pays for, on business terms, usually comes with a processing agreement. A personal, free or consumer account usually doesn't. Our article Who's who in your AI supply chain? discusses this case.

Through AI Integrations on Pro or Enterprise. Replit's agreement names Replit as the processor and gives it a general authorisation to use sub-processors. Replit's sub-processor list (August 2026) names OpenAI, Anthropic and Google ("AI models") and OpenRouter ("AI model selection for Integrations"), all in the USA.

But we didn't find anywhere in Replit's documents that says your app's AI calls fall under the DPA. The agreement's own term for what it covers, "Customer Data", isn't defined either. The Pro and Enterprise agreements also say third-party services "may be subject to additional or separate terms", and that where those conflict with the agreement, "the Third Party Terms shall control". We didn't find a document that says whether AI Integrations counts as one. Three possibilities are open:

Through AI Integrations on Core. You don't have a processing agreement with Replit, or an account with the model's maker. That doesn't make Replit a controller, or a processor: the roles follow what each company does. It means the contract the law expects between you and anyone handling the data on your behalf (GDPR Article 28(3)) is not in Replit's Core terms.

The third possibility above may apply on Core too. The Terms of Service reserve Replit's access to the content of private apps for, among other things, "improving our service", and whether that reaches your users' data when the app runs isn't stated. The sub-processor list says it covers processing under the Terms of Service as well as the commercial agreement, so a Core builder can see the same names. It doesn't say which of the model makers on it receive AI Integrations traffic.

So what should you read? Check your plan, then read the platform's published processing agreement, its documentation for the AI features and its sub-processor list. On 21 October 2026, we didn't find an answer in those three to whether AI Integrations calls are covered. That may mean there isn't one, or only that we missed it.

Boundsmap, the tool we make to map who is responsible for what in an AI product, asks about this: how your product reaches the model. If you reach it through a cloud platform rather than from the model's maker, or build on a platform that calls the model for you, that platform is the one you contract with.

A note on free models

Replit's page says that for self-serve users, OpenRouter requests are sent with "Free endpoints training enabled: Free model providers may train on your prompts and completions" and "Free endpoints publishing enabled: Free providers may publish your prompts and completions to public datasets". Paid endpoints have training disabled.

If your app sends users' data to a model that can keep it or reuse it, that is part of the picture of who handles your users' data. If your users' data ends up published or reused that way, who is responsible: you, Replit or OpenRouter? We don't think the answer is clear.

What Replit says you're responsible for

Replit's shared responsibility page, read on 21 October 2026, lists "Regulatory compliance of your application" (its examples include GDPR) as yours: the table's responsibility column says "You". That's just how Replit divides the work in its own documents - it's not the law. A company that handles personal data on someone else's behalf still has obligations of its own.

The same table says Replit vets and discloses subprocessors that process customer data on the platform, and that you vet and disclose subprocessors used by your app. It doesn't say whether the AI models your app uses through AI Integrations count as Replit's subprocessors or yours. The model makers are on Replit's list, and your app's AI calls go to them too.

So, whatever your plan, these are the questions to ask about your own position:

Additionally, every company named above is listed in the USA. Replit's transfer clauses are in its DPA, so on Core there are none in its terms. Sending personal data there is an international transfer with rules of its own, which our Who's who article covers.

If your customers are businesses

Where your app is used by other companies on their own clients' data, the personal data is theirs and you handle it on their behalf. Replit's agreement provides for that case on Pro and Enterprise.

Whether you must name OpenAI, Anthropic, Google or OpenRouter to those customers, or whether naming Replit is enough, isn't settled. EU guidance leans towards the customer having the full chain to hand.

And the EU AI Act?

The EU's AI Act can reach an app built anywhere, if it's offered to people in the EU or its output is used there (Article 2(1)). Whether you count as the provider of your app's AI is a separate question, and the AI Act doesn't answer it for every system. Our Who's who article sets out what is settled and what is not.

If your app talks to people, as a chat app does, the AI Act asks that they are told they are dealing with an AI (Article 50(1)), and that AI-generated text, images, audio or video is marked in a machine-readable way (Article 50(2)). Both have applied since 2 August 2026. An app placed on the market before then has until 2 December 2026, and only for the marking. The notice isn't needed where it's obvious to a reasonably well-informed person, given the context, that they're dealing with an AI. The marking isn't needed where the AI only assists with standard editing, or doesn't substantially change the text it was given. The text of Article 50 and the Commission's guidelines on it have the details.

Boundsmap

Whatever Replit plan you're on, you're left to work out your own position: who handles your users' data at each step, on whose terms, and whose obligation each step is. That's easier to see as a picture than as a list, and it's why we built Boundsmap.

It maps your AI product onto five layers and four boundaries, and works out what the law asks in the UK and the EU at each one, and whose obligation it is. You describe your system in plain words. The rules engine is deterministic and tested, and it isn't an AI model. Where your answers leave something open, Boundsmap asks a question you can answer without knowing the law, such as how your product reaches the model. Where the law itself is unclear, Boundsmap says so and shows the different ways it could be read.

It's important to note that Boundsmap doesn't give legal advice and doesn't replace an adviser. It gives you a clear, checkable map of your own position, and a place to start the conversation with your team, your customers and a lawyer.

You can map your own product. No signup is required, but expect to spend about 10 to 15 minutes on it. Or see an example map of a chat assistant whose company reaches the model through a cloud platform.

Before publishing, we asked Replit a few questions about their data processing agreement and AI Integrations, and we're grateful for their reply. They pointed us to the documents they recommend citing: their Data Processing Agreement, AI Integrations documentation and sub-processor list. Those are what this page quotes.