Mozilla has launched Smart Window, a browser assistant built on Mistral's models, and the two companies are positioning it as AI-powered browsing that puts privacy first. The assistant is available first in France and North America, with the UK and Germany to follow later this year. The choice of a European model provider rather than a US hyperscaler is the part that matters for companies now weighing which AI vendors they let near employee browsing data.
What Smart Window does
The assistant handles three tasks: complex searches, remembering content a user has already visited, and summarizing information from open tabs. That combination moves the browser from a passive window into a working memory layer — the user does not have to re-find a page or copy text between tabs to assemble an answer. Availability is staged by geography rather than by product tier: France and North America first, the UK and Germany later in the year. Mozilla has not published pricing, seat limits, or enterprise administration terms in the announcement.
The privacy claim rests on Mistral committing to store no data. That is a contractual commitment from the model provider, not a technical description of where inference runs or what the browser keeps locally, and the two are not the same thing. For a business, the distinction decides whether the tool can be used on client material at all: a no-storage pledge from the vendor covers the model side, while retention on the device or in the browser account is a separate question. Mozilla frames the browser as a place where different AI providers should be able to compete and where open source keeps a firm spot — which implies Smart Window is meant as one option among several rather than a single locked-in assistant.
The partnership also marks a shift in Mistral's commercial direction. The company has built its position on business customers, and reaching end users worldwide through a browser shipped by Mozilla is a different distribution channel from selling API access to companies. Mozilla, for its part, has spent years arguing that the browser is the natural control point for how people meet AI, and it now has a named model partner to test that argument in two large markets before the UK and Germany are added. Both moves follow the same logic: the assistant layer is being contested by browser makers and model providers at the same time, and each side needs the other to reach users.
What this means for business
For companies that already allow AI assistants in the workplace, the practical change is the arrival of a second sourcing model. A browser assistant tied to a European provider gives procurement an alternative to assistants bundled with US cloud contracts, and the no-storage commitment is the kind of clause legal teams ask for when browsing history may contain customer names or unreleased plans. The effect is most visible in small companies, where a browser extension is often the only AI tool employees actually adopt, and in large ones, where the same feature has to pass a vendor review before it reaches a single desk.
What the announcement does not settle is how the assistant behaves under enterprise control. There is no information on whether administrators can disable it centrally, whether visited-page memory can be scoped per user or per device, or what happens to stored context when an employee leaves. The no-data-storage commitment applies to Mistral; it says nothing about Mozilla's own retention. Before adopting, the questions to put to the vendor are concrete: where inference runs, what the browser keeps after a session ends, whether the assistant can be turned off by policy, and how the no-storage pledge is verified rather than stated.
The marker to watch is the UK and Germany rollout later this year. If the assistant ships there on the same terms, with the no-storage commitment intact and enterprise controls published, browser-based AI becomes a realistic default for regulated teams rather than an experiment in two markets. If the expansion slips or the terms change, the privacy-first framing stays a positioning claim, and companies should keep treating the browser as a place for consumer tools, not for work that touches client data.
