AI assistants
Lumo privacy rating
AI assistant from Proton that runs open-weight models on Proton's servers, with no logs of chats and zero-access encrypted chat history.
Summary
Lumo scores 73 out of 100 (grade C) on the AI assistants criteria. It meets 7 of 12 criteria: no ads or data sales, transparency report, tells users about requests, TLS configuration, security headers, no training on your data and limited chat retention. It partly meets open source, no trackers or telemetry, independent audit and no account needed. It does not meet runs locally. It is based in Switzerland: Not in the Five, Nine or Fourteen Eyes; GDPR-style data protection law. Automated tests: SSL Labs grade A+ and Mozilla HTTP Observatory grade A+.
Score 73 out of 100. How scoring works
Criteria
-
Partial
Open source Weight 3 of 3
Is all the source code needed to run the product public?
Apps are open source under GPL-3.0. The server is not.
-
Partial
No trackers or telemetry Weight 3 of 3
Are the website and apps free of third-party trackers, with any analytics anonymous and any telemetry off by default?
Website analytics are self-hosted, and the Android app includes Sentry crash reporting.
-
Yes
No ads or data sales Weight 2 of 3
Is the product funded without advertising, ad targeting or selling user data?
Funded by paid plans, with no ads.
-
Partial
Independent audit Weight 2 of 3
Has an independent security or privacy audit been published within the last three years?
Proton completed a SOC 2 Type II audit, but the report is not public.
-
Yes
Transparency report Weight 2 of 3
Does the provider regularly publish how many government and legal requests it receives and how it responds?
Publishes yearly counts of legal orders received, complied with and contested.
-
Yes
Tells users about requests Weight 1 of 3
Does the provider promise to tell users about requests for their data, unless a court forbids it?
Targeted users are notified of data requests, with delays only when Swiss law, a court order or a risk to life requires it.
-
Yes
No training on your data Weight 3 of 3
Are prompts, chats and files kept out of model training by default?
Chats are not used to train models, except anonymized feedback that users choose to share for one model.
-
No
Runs locally Weight 2 of 3
Can the assistant run on your own device or server, so prompts never leave it?
Hosted only.
-
Yes
Limited chat retention Weight 2 of 3
Are chats deleted by default or on request, with no long-term server copy?
Chats are erased from servers after each response. Saved history is zero-access encrypted and only readable by the user.
-
Partial
No account needed Weight 1 of 3
Can it be used without an account or personal details?
Guest access works without an account, with usage limits.
Automated tests
-
Yes
TLS configuration Weight 2 of 3
Does the website pass the Qualys SSL Labs test with a grade of A or better?
Grade A+
-
Yes
Security headers Weight 1 of 3
Does the website pass the Mozilla HTTP Observatory test with a grade of A or better?
Grade A+ (115/100+)
-
Not tested yet
Modern web standards Weight 1 of 3
Does the website score 90% or higher on the Internet.nl website test?
Not tested yet.