Skip to main content

Can Attorly keep AI processing in the EU? What runs where

Last updated:

Short answer

Yes, if you turn it on. By default Attorly’s AI runs with US-based model providers under EU standard contractual clauses. Two options in Settings → Privacy, free on every plan, change that: EU-only AI processing runs chat, document analysis, drafting and text recognition on Mistral Large, processed in the EU, and EU embeddings indexes your documents with Mistral Embed in the EU. With both on, no content of yours goes to a model provider outside the EU. Zero data retention is not available yet. The application and its database run in the EU, and uploaded files are stored in a Cloudflare R2 bucket under Cloudflare’s EU jurisdiction.

What runs where

Part of AttorlyBy defaultWith EU-only AI processingWith EU embeddings
Chat, document analysis, drafting and research answersThe model you choose; the default is Anthropic Claude (United States). OpenAI (United States), Google Gemini (global) and Mistral (EU) can be chosenMistral Large only, through Attorly’s AI gateway (Netherlands) to Mistral (EU). Features that would use another model run on Mistral LargeNo change
Text recognition (OCR) of scans and imagesMistral OCR (EU)Mistral OCR (EU)No change
Indexing your documents and saved items for searchOpenAI embeddings (United States)No changeMistral Embed (EU), through the gateway
Meaning-based search in legal sources such as Lovdata and EUR-LexYour query is embedded with Google (global) and combined with keyword searchNo changeKeyword search only, because the legal-source index is built with a non-EU model
Background jobs that run as provider batchesAnthropic or OpenAI batch APIs (United States)Each item runs one at a time on Mistral Large insteadNo change

The options are enforced on the server: every call that carries your content is checked against them before it leaves Attorly, and a call that would break an option is refused or moved to a model that meets it. Nothing falls back to a non-EU provider.

Where your data is stored

  • Application and database: Railway’s EU-West region.
  • Uploaded files: encrypted in the application, then stored in a Cloudflare R2 bucket under Cloudflare’s EU jurisdiction.
  • Database backups: Cloudflare R2 buckets with a Western Europe location hint, which is a placement preference, not an EU jurisdiction.
  • Cloudflare and the AI providers are US companies; transfers rely on Standard Contractual Clauses or the EU-US Data Privacy Framework, as listed on the subprocessors page.
  • Enterprise can bring its own storage bucket and encryption keys (AWS KMS or Azure Key Vault) to keep files in a region it chooses.

Zero data retention: not available yet

The zero data retention option sends your content only to providers that have agreed in writing to keep none of it once they have answered. None of the chat providers Attorly uses has signed such an agreement with us yet, so the switch is shown as not available. It becomes available, and this page will say which providers it covers, when one does.

Mistral’s OCR API keeps no content by default for every customer. Provider batches are never zero-retention, because the gateway keeps batch results for up to 30 days. Every provider is used under API terms that exclude training on customer content.

Who can turn the options on, and what it costs

  • Any user on any plan, for themself, under Settings → Privacy. Each option is separate and off by default.
  • A workspace owner or admin, for every member, in the workspace settings. A member cannot turn off an option the workspace has turned on, but can add more for themself.
  • EU-only AI processing costs nothing extra. Mistral Large uses fewer AI tokens per answer than the default Claude models.
  • EU embeddings costs more than standard indexing: about 1,300 AI tokens for a 20-page contract instead of about 260, plus a few tokens per search. Turning it on re-indexes your existing documents in the background, billed to your AI token balance like any indexing.

What you give up

  • Model choice: with EU-only AI processing you work with Mistral Large, not Claude, GPT or Gemini.
  • Legal-source search: with EU embeddings, search in Lovdata, EUR-Lex and the other legal sources matches keywords only, without meaning-based retrieval. Answers still cite the sources.
  • Saved clauses, research notes and other saved items move to the EU index the next time each is saved; until then they do not appear in meaning-based search.

Settings → Security → Data residency shows the live picture for your account: where storage runs, which AI providers and regions are configured, and which options are on.

Frequently asked questions

Does Attorly have EU data residency?
The application, the database and uploaded files are in the EU. AI processing is in the EU only when you or your workspace turn on EU-only AI processing and EU embeddings; by default it runs with US-based providers under EU standard contractual clauses. Database backups are not under an EU jurisdiction.
Is client data used to train AI models?
No. Attorly does not train models on your content, and every AI provider it uses is bound by API terms that exclude training on customer data.
Can Attorly guarantee zero data retention at the AI provider?
Not yet. No chat provider Attorly uses has signed a zero-retention agreement with us, so the option is shown as not available. Mistral OCR keeps no content by default.
Which plans include EU-only AI processing?
All of them. The options are free to turn on; only EU embeddings uses more AI tokens than standard indexing.
Can we run the models ourselves?
Yes. Pro and Enterprise can run analysis on a self-hosted Mistral deployment, reached over HTTPS only; its region and retention are then yours to decide.

Try it with EU-only processing from day one

Start a 7-day trial, turn on EU-only AI processing and EU embeddings under Settings → Privacy before you upload anything, and check the result under Data residency.

Start your trial