Meet Seal AI: The Analytics Assistant That Never Sends Your Data to the US
Ask your analytics a question in plain language and get a grounded answer — from an AI that runs in the EU only, keeps nothing, and trains no one's model. Private by architecture, not by promise.
Key Takeaways
- Seal AI answers analytics questions in plain language by querying your real data — grounded in your numbers, not invented.
- Inference runs in Paris only, on a European provider with no US parent. No data leaves the EU, so no international transfer ever happens.
- Zero prompt retention, no training on your data, and an open-weight model — privacy is a property of the architecture, not a policy layered on top.
- No setup: no API keys, no AI vendor account. Prefer your own provider? Bring your own key — but Seal AI is the default so privacy isn't optional.
Analytics tools have been racing to bolt an AI chat box onto the dashboard. Most of them quietly route your data to a US AI provider to do it. We didn't want to be one of them — so we built Seal AI, the private AI layer inside SealMetrics, on a different premise: the AI should be as private as the analytics it explains.
What it does
Seal AI is an analyst you can talk to. Ask "how did traffic do versus last month?" or "which of my top sources converts best, and why do you say that?" and it answers in plain language — by actually querying your data, not by guessing. Under the hood it plans, calls the right data tools, reads the results, and writes them back as a clear answer with the occasional chart or table. It also produces automated insights on a schedule, so patterns surface without you asking.
Crucially, it's grounded: it narrates the numbers it retrieved from your account and is built to refuse to invent ones it didn't. We test that discipline continuously against ground truth pulled straight from the database.
Why it's private by architecture
"Private AI" gets used loosely, so here is exactly what we mean, and it's all structural:
- EU-only inference. The model runs on Scaleway's infrastructure in Paris, and only there. Scaleway is a French company with a French parent and no US ownership, so it isn't subject to the US CLOUD Act.
- No transfer, so no transfer problem. Because your data never leaves the EU and the provider has no US parent, GDPR Chapter V simply isn't triggered — no Standard Contractual Clauses, no transfer assessment, no dependence on the EU-US Data Privacy Framework surviving its next court date.
- Zero retention, no training. Scaleway retains no prompt content by default; SealMetrics stores only token counts, never the content. The model is open-weight and static — it does not learn from your requests.
- The prompt is born clean. SealMetrics is consentless analytics: we never collect IPs, cookies, or visitor identifiers in the first place. So there is no personal identifier to send to the AI, even before privacy controls apply.
The difference from bolting on a US AI API is not a stricter contract. It's that the data never enters a jurisdiction where a stricter contract would be needed.
No setup, and no lock-in
You don't configure anything. There are no API keys to create and no third-party AI account to sign up for — Seal AI is included and uses a platform-owned key you never see. If you'd rather use a specific external provider, SealMetrics supports bring-your-own-key for OpenAI, Anthropic, Gemini, or DeepSeek. But Seal AI is the default on purpose: privacy shouldn't depend on a checkbox.
And because the model is open-weight under the Apache 2.0 license, we're not locked to a single vendor either. The same model is available from several European hosts and can run on a single GPU — so the privacy guarantee is one we could keep even if we changed infrastructure. A guarantee you can't walk away from isn't a guarantee; ours is portable by design.
Try it with three questions
If you already use SealMetrics, open the assistant and ask: (1) "Compare my entrances over the last 7 days to the previous 7." (2) "Which of my top 3 sources converts best, and why?" (3) "What's my bounce rate this month?" You'll get grounded answers in seconds — and none of the data behind them will have left the EU.
Want the engineering detail? We've published how Seal AI works, end to end, including the data-handling and compliance analysis, in our technical documentation.
Frequently asked questions
What is Seal AI?
Seal AI is the private AI layer built into SealMetrics. It powers a natural-language assistant that answers questions about your analytics by querying your data, and it generates automated insights. Inference runs on European infrastructure only (Scaleway, Paris), retains no prompt content, and never trains any model on your data. It requires no setup: there are no API keys to create and no AI vendor account.
Does Seal AI send my data to the United States?
No. Seal AI runs inference exclusively in Paris, France, on Scaleway — a French company with no US parent. Because the data never leaves the EU and the provider is not subject to US jurisdiction, no international transfer occurs, so GDPR Chapter V (Standard Contractual Clauses, transfer assessments, the Data Privacy Framework) does not apply.
Does Seal AI train on my analytics data?
No. The underlying model (gpt-oss-120b) is open-weight and static — it does not learn from requests. Neither Scaleway, nor SealMetrics, nor the model's creator trains on your data. SealMetrics stores only token counters for quota and billing; prompt and response content is never persisted.
How is Seal AI different from using ChatGPT or Claude with my data?
Consumer and API AI tools from US providers route your data to US-jurisdiction companies, even when an EU region is available. Seal AI is private by architecture: EU-only inference, no US parent, zero retention, no training, and an open-weight model you could self-host. If you prefer a specific external provider, SealMetrics also supports bring-your-own-key — but Seal AI is the default so privacy doesn't depend on configuration.
What can I ask Seal AI?
Anything you'd ask a data analyst about your traffic and conversions: how a period compares to the previous one, which channel or source converts best and why, what your bounce rate is, how devices break down, or why a metric moved. It answers by calling your real data tools and narrating the result — grounded in your numbers, not invented.