Evanesca vs Lumo by Proton: two private AI chats, one Swiss and one French

At a glance

Each Lumo cell links to the Proton page it comes from, read on 6 October 2026; each Evanesca cell links to the page that describes it.

How Lumo protects your chats

« After Lumo processes your query and generates a response, the data is erased. »

Proton describes several layers source. Prompts travel over TLS and are also encrypted so that only Lumo's GPU servers can decrypt them. Lumo keeps no logs of what you ask or what it replies, and stores no metadata such as timestamps or IP addresses. Your history is kept on your device and, on the Free and Plus plans, synced to Proton's servers with zero-access encryption, so only you can read it; as a guest, the conversation is erased at the end of the session.

Lumo does not use your chats to train models; the one exception is anonymised feedback on Apertus 1.5 answers, sent only if you choose to. And Lumo's code is open source, so these claims can be checked source.

What Evanesca keeps, and what it does not

Evanesca makes the same promise on the content: your prompt and the answer are held in memory while the answer is written, never stored, and never used by Evanesca to train a model. Your history lives in your browser — on that device only, with no synced copy.

It keeps more than Lumo around the content: billing data for each answer, tokens and cost, and your account and payment records as the law requires. The operator is named, and states its own rules: OVHcloud keeps “only the data required for billing purposes” source; Scaleway may keep and read requests that cause errors or look abusive, for up to 14 days source; Mistral AI keeps its models' inputs and outputs 30 days to monitor abuse source; Infomaniak, which runs Apertus, states that queries are “neither recorded nor used to train models” source.

In exchange, each answer carries a passport: the model that wrote it, the operator that ran it, the country, the law, and what was kept. Lumo chooses the model for you; Evanesca shows you which one answered.

Swiss law, French law

Lumo is run by Proton AG, a Swiss company in Geneva overseen by the non-profit Proton Foundation; for consumers outside the United States, its terms are governed by Swiss law and the courts of Geneva source. Proton says Lumo runs on its “fully European infrastructure, protected by Swiss privacy laws and zero-access encryption” source; it does not name the countries of its servers on these pages.

Switzerland is in Europe but not in the European Union. The European Commission recognises Swiss data protection as adequate, so data can flow from the EU without extra safeguards source. Neither service is run by a US company.

Most of Evanesca's models run in France, with OVHcloud or Scaleway, French companies with no US parent, under French and EU law; Mistral AI runs some of its own models in its EU region, AKI.IO runs others in Germany, and Infomaniak runs Apertus in Switzerland. If your policy asks for processing inside the EU itself, each model's page and each answer's passport give the country, so you can keep to the models run in the EU sovereignty · AI Act.

The models behind each

Lumo's text models are Qwen 3.5, GLM 5.3 and Apertus 1.5, the Swiss open model, plus two image models; they run only on servers Proton controls, and Lumo routes each request to “the model most capable of dealing with your request” source.

Evanesca lets you choose the model, or leave it to Auto mode, which picks among the models run in Europe; you can switch model mid-conversation. Its catalogue includes Qwen3.5 and GLM-5.2, run in France, and Apertus v1.5 70B, run by Infomaniak in Switzerland. GLM 5.3, which Lumo runs on Proton's own servers, is not in Evanesca's catalogue: Evanesca serves only models run in Europe, and no European operator runs it for Evanesca. Compare Evanesca's models.

Where Lumo is the better choice

• Encrypted history synced across your devices, which only you can read source.
• No metadata stored, and use as a guest without any account source.
• Open-source code, so its privacy claims can be audited source.
• Image editing, files linked from Proton Drive, and apps for iOS and Android source · source.
• A free plan, part of a wider suite of private services from the same company; and it runs today, while Evanesca has not opened yet source.

If encryption and sync across devices matter most to you, and Swiss jurisdiction suits you, Lumo is a good choice.

Where Evanesca fits better

• Processing in Europe — in France for most models, by French operators, under EU law sovereignty.
• An Agent mode that carries out a whole task — it searches the web, writes Word, PDF and Excel documents and creates images — within a spending cap you set Agent mode.
• A choice of models from several makers, each showing its checks and priced in public, and the freedom to pick one per answer models · benchmarks.
• A passport under each answer that records the model, operator, country, law and what was kept passport.
• No subscription: you pay for the answers you get, VAT included, and credit never expires pricing.
• An OpenAI-compatible API for the same models, and for teams a signed Art. 28 DPA docs · business.

Evanesca also reads documents and images, draws charts, handles code and maths in the browser, and takes dictation why Evanesca.

How to choose

• Private chat on several devices, with no account if you like, and no metadata at all: Lumo.
• Work data that must be processed in the EU, with a record of which model, operator and country handled each answer: Evanesca.
• Tasks to hand over — research, a report, a spreadsheet — with a spending cap: Evanesca's Agent mode.
• Developers who want the models behind an API, paid per token: Evanesca.

The two are not rivals on principles: both treat your conversations as yours. They answer different requirements — encryption and sync on one side, EU processing, model choice, tasks and per-answer records on the other.

How this page was checked

Proton's pages were read on 6 October 2026 and are quoted in their own words. Lumo's models change over time — Proton lists the current ones on its privacy article — so check that page before relying on a model name. If a line is out of date, write to us.

• Proton — Lumo privacy
• Proton — about Lumo
• Proton — Lumo pricing and plans
• Proton — Lumo 2.0 announcement, 30 June 2026
• Proton — terms of service, last modified 23 June 2026
• European Commission — adequacy decisions

Frequently asked questions