TRUST & TRANSPARENCY
Trust is earned in the open.
Privacy products shouldn't ask for faith. Here's exactly what we do and don't do, in terms you can check yourself.
What leaves your device
The short version: almost nothing. Here's the full accounting.
Your prompts & files
Processed on-device. Sensitive values are replaced with one-way placeholders before anything is sent to YOUR configured AI provider. Reversible tokenization is planned shortly after Locke 1.0.
Detection results & counters
Stay in local files on your machine, with owner-only permissions.
Telemetry / analytics
No behavioural analytics, and no background channel for it. Crash reports and support messages exist but only travel if you send them.
Your data to Sonomos servers
Nothing. We don't have servers that receive it.
This website
Static pages. No cookies, no tracking — just anonymous analytics to keep things running. See our no-tracking policy and our website privacy policy, which covers every site on our sonomos.ai domain. Locke itself is covered by a separate privacy policy.
How we make money
Canary is free because Locke is the business. No ads, no data resale, no "free product where you're the product." We don't sell data and we have no data to sell — the processing that matters happens on your machine, not ours.
Check our work
Don't take "zero network requests" on faith. Read the code, or watch the CI enforce it.
Canary is MIT-licensed and open source
Read every line, run it yourself, or fork it.
github.com/sonomoshq/CanaryZero network requests, enforced mechanically
"No network requests" isn't a claim in a README — it's a check CI runs on every commit, not a promise we're asking you to believe.
The Sonomos Trust Center
Locke's security documentation for enterprise review — architecture, threat model, compliance readiness, and known gaps, disclosed openly. Access is granted for security reviews.
trust.sonomos.aiWhere we are, honestly
Our engineering repos keep an HONEST.md file listing what isn't true yet. We'd rather under-claim than let a gap surprise you later — so here's the same honesty, in public.
Locke is pre-1.0. It hasn't yet undergone a formal third-party security audit or a SOC 2 Type II audit. Both are planned.
Our claims are backed by reproducible benchmarks, not certifications — for now. You can run the same benchmarks we do, straight from the repos.
Report a vulnerability
Found a security issue? We welcome responsible disclosure and commit to responding. See our security.txt for the formal policy.
Ready to see it for yourself?