Legal

Privacy Policy.

CostMyAI ("we," "us," "our") is operated from Vienna, Austria. This policy explains what data we collect, why, and what rights you have over it.

Last updated: 20 August 2026

  • What we do not collect

    We do not collect, store, or have access to your AI provider API keys or credentials. The Verification Engine that reads your usage data runs inside your own environment. Your provider keys never leave it.

  • What we do collect

    Account data: name, email, workspace details, and billing information necessary to provide the service.

    Usage and billing data you choose to share: aggregate, provider neutral records of model usage, spend, and switching activity, pushed from your own environment via the Verification Engine. Model outputs are never sent to us, and prompt content is only ever involved in the one opt-in case described immediately below. Which Verification Engine image you run decides which labels you get, and the quickstart hands new customers the v3 image by default. On the v1 image, the task label is derived from the request path and model name alone and no request body is read. On the v2 and v3 images, local classification is on by default: request text is read inside your own container to derive a more accurate label, and COSTMYAI_CLASSIFY_LOCAL=false turns it off. On all three images, only the label, a confidence number and feature names leave your environment by default, and no prompt text is transmitted, stored or logged by us.

    The one exception — opt-in only, on the v3 image: the v3 Verification Engine image is capable of remote classification, and it is off by default, exactly as on v2. Only if you explicitly set COSTMYAI_CLASSIFY_REMOTE=true does it activate: when the in-container rules cannot place a request, the extracted request text is sent to us and labelled by a model. That call runs after your response has already been returned, so it never delays your traffic. The text is used for that single labelling call and is not persisted by us; the only thing retained is the resulting label. Left unset, the v3 image behaves exactly like v2, where the capability does not exist at all; the v1 image reads no request body whatsoever. Both quieter tags remain published and supported, and your own environment variable always overrides the tag's default. Where remote classification is opted into, OpenAI processes the extracted text as a subprocessor — see the subprocessor list below.

    Website and product analytics: we use Google Analytics to understand how visitors use this site (pages visited, general location, device and browser type, session behavior). Google Analytics sets cookies and processes data including your IP address, per Google's own IP handling settings, and device identifiers. This data is processed by Google in its own role as a processor on our behalf, and is transferred to Google's servers, which can include servers outside the EU/EEA.

  • Cookies and consent

    Google Analytics is not strictly necessary for the site to function, so under GDPR it may only run after you have given consent, not merely with an opt-out available afterward.

    You can withdraw consent at any time through your cookie settings, or by using your browser's own cookie controls.

  • Why we process it

    To provide the service you signed up for, to bill you accurately, to respond to support requests, and where you have consented, to contact you about product updates.

  • Legal basis (EU/EEA users)

    We process personal data on the basis of contract necessity (to provide the service), legitimate interest (product improvement, security), and consent (marketing communications, where applicable).

  • Data sharing

    We do not sell personal data. We share data with subprocessors strictly necessary to run the service and the site, each bound by their own data processing agreement:

    • Google (Google Analytics): website analytics, per the Cookies and consent section above.
    • OpenAI: only when a customer explicitly opts a v3 Verification Engine container into remote classification with COSTMYAI_CLASSIFY_REMOTE=true. In that case the extracted request text the in-container rules could not place is sent to OpenAI for a single labelling call, and is not retained by OpenAI or by CostMyAI beyond that call's response.
    • Stripe: payment processing and subscription billing.
    • Our hosting and database providers: operation of the application and storage of account and usage records.
  • How your usage data is used across service tiers

    CostMyAI sells four tiers: Compare, Certify, Rightsize and Govern. The usage data you share is used to compute recommendations for all four, regardless of which tier you are subscribed to, including tiers you have not bought and tiers that are not yet generally released.

    We do this for one reason: a recommendation engine can only be validated against real usage. Computing the higher-tier results on live data is how we check that they are correct before those tiers go on general sale.

    Results you are not entitled to are computed and stored, but they are not shown to you. The database itself enforces this: a workspace can only read a recommendation whose tier is at or below the plan it actually pays for. If you later upgrade, the results for that tier become visible to you; if you never upgrade, you never see them. These computed results are your workspace's data, are covered by the same retention and deletion rules below, and are never shared with, sold to, or used to advise another customer.

  • Data retention

    We retain account and billing data for as long as your account is active and as required by law afterward. Usage data pushed via the Verification Engine, and everything derived from it, follows the retention rules set out in the Retention section of our Methodology page: kept while your workspace exists, deleted within 30 days of closure or on request.

  • Your rights

    If you are located in the EU/EEA, UK, or another jurisdiction with equivalent protections, you have the right to access, correct, delete, or export your personal data, and to object to or restrict certain processing. Contact mail@costmyai.com to exercise any of these rights.

  • International transfers

    Google Analytics can transfer data to Google servers outside the EU/EEA. Google is certified under the EU-U.S. Data Privacy Framework and relies on Standard Contractual Clauses for such transfers, per Google's own published terms. Any other subprocessor outside the EU/EEA operates under an equivalent transfer mechanism.

  • Changes to this policy

    We will notify active accounts of material changes to this policy before they take effect.