∴
Privacy & safety

What we promise. What we don't.

A premise has to stand up to scrutiny. These are the commitments the product is built on — plain language, no fine print that contradicts it.

∴
What we commit to
  1. 1.
    We never train on what you type.
    Your prompts aren't used to train any model — ours or anyone else's. Not now, not in any future pivot. There is no pipeline that does this.
  2. 2.
    We don't store the content of prompts.
    Each prompt is processed in memory and discarded once the critique is generated. Text is never written to our database. What we keep is which rules fired — never what triggered them.
  3. 3.
    We process locally when we can.
    The extraction step runs on your device when hardware supports it. When it doesn't, we route to a named provider under a published no-retention agreement. The app shows you which mode it's running in.
  4. 4.
    No third-party tracking.
    No Facebook pixel, no Google Analytics, no session replay, no ad SDKs — on this site or in the app.
  5. 5.
    We don't sell, share, or rent data.
    Not anonymized. Not aggregated. Not "for research." If this ever needed to change, the product would shut down first.
  6. 6.
    The rule catalogue is public.
    Every check we run, with the reason we run it, lives at /rules. If it isn't there, we don't run it.
  7. 7.
    Domain experts review safety-critical rules.
    Health, FemTech, mental health, and any future regulated domain get external review by a credentialed clinician before launch, and quarterly after. Reviewer names are on the catalogue.
  8. 8.
    Organizations get the same defaults, plus controls.
    For teams: SSO, audit log, regional data residency. Admin tools show coverage statistics, never employee prompt content. Healthcare orgs can sign a BAA. Everything above applies by default to every personal account.
∴
What we don't promise
  1. ·
    We don't make AI safer.
    We help you ask the right question. The model still answers — and models can still be wrong, biased, or out of date. We close one specific gap. Not all of them.
  2. ·
    We don't replace professionals.
    WithContxt helps you think before you ask AI. It does not replace a clinician, lawyer, accountant, or therapist. When the question warrants one, we say so.
  3. ·
    We don't catch every bad question.
    The catalogue grows, but it won't ever be complete. If you find a gap, send it — that's how the catalogue gets better.