Privacy Policy
Last updated: November 11, 2026
1. Introduction
GPU Finder (gpufinder.dev) is a GPU cloud price comparison tool operated by Alexander Salikov as an independent project. This Privacy Policy explains what data we collect, how we use it, and your rights regarding that data.
2. Data We Collect
Subscriptions and Email Addresses
If you sign up for price alerts or product updates, we collect your email address and the preferences needed to provide that subscription. For price alerts, this can include the GPU, quantity, price threshold, and delivery preferences you selected. We also temporarily store the IP address used to subscribe for abuse prevention. The API early-access waitlist is retired; existing waitlist records and unsubscribe links are retained under the subscription retention rules below.
Provider Requests
If you request that we add a cloud provider, we collect the provider name and any provider URL, notes, or email address you choose to submit. We also store the request IP address to prevent spam.
Developer Accounts and API Access
If you use the developer portal, we store your verified email address and account details. Email sign-in uses a single-use email link; the link token is stored as a hash and expires after 10 minutes. A signed-in browser receives an HTTP-only session cookie. A server-side session record includes its expiry and may also include the session IP address and browser user-agent for account security. API keys have a label, visible prefix, status and lifecycle dates. The full key is shown once when created or rotated; we store its SHA-256 hash for authentication, not the raw key.
When Google sign-in is enabled and you choose it, Google shares your identity, email verification status, name and profile image. We store the Google account identifier to connect it to your developer account, and may store your name and profile image when creating that account. Google access, refresh and ID tokens are discarded after authentication. We request only identity information, with no access to Gmail or Drive. Connecting Google can require a fresh email confirmation; the confirmation record expires after ten minutes. Recovery challenges are stored as hashes and expire after ten minutes. Support recovery keeps a hold and review receipt, revokes old credentials and requires a new email sign-in. Recovery records remain as needed to protect the account and document the review.
We also store account-linked API usage and rate-limit counters, access and key-management audit records, and any allowance-extension request and its review. These records support the API, account controls, billing and abuse prevention; audit records are designed not to contain API keys, email addresses or request payloads.
If you connect an MCP application with OAuth, we store its registration, your consent and granted scopes, token hashes and expiry dates, linked to your account and sign-in session. The application receives read-only GPU data and account usage through the permissions you approve. You can disconnect it in the developer portal to stop access and token renewal. Disconnecting preserves account usage and security audit records.
Data Returned to Connected MCP Applications
When you use GPU Finder through an assistant such as ChatGPT, Codex or Claude, the connected application receives the GPU and provider information you request. Responses also include your account's monthly and per-minute usage totals, limits, remaining allowance and reset times so the application can respect those limits. A charged call includes its operation name and unit cost. Errors can include an error category, a safe explanation, retry and reset guidance, accounting outcome and recovery links. Paginated results include an opaque continuation cursor. Data freshness timestamps describe the GPU catalog, not your personal activity.
These disclosures cover nested response fields and both structured and text tool results, including error responses. MCP response bodies do not include your email, name, internal account or credential identifiers, billing identifiers, receipt identifiers, internal request identifiers, raw collector errors or debug records. A server-generated request correlation identifier may be included in HTTP headers for support and is retained in operational logs. The connected application processes and retains the information it receives under its own privacy policy. You control which calls you make and can revoke OAuth access or your API key in the developer portal.
Billing
If you start a paid plan, Stripe processes payments through its hosted Checkout. When payment-method management is configured for your account, the developer portal can send you to Stripe's hosted page to update payment details. We store Stripe customer, checkout, subscription and event identifiers, along with plan, amount, currency and status information needed to reconcile access. We do not store payment-card numbers in GPU Finder. Stripe may process billing information under its own privacy policy.
Outbound Provider Links
When you follow certain links from GPU Finder to a cloud provider, we record the provider identifier, a shortened referrer value, and shortened browser user-agent information. We use this to understand whether provider listings are useful and to detect abuse.
Error and Performance Monitoring
We use Sentry to diagnose application errors and slow requests. Reports include exception details, stack traces, application version, environment, sampled performance timings, and selected operational logs about job health and service failures. These logs use predefined fields rather than request or payment payloads. Automatic collection of account identity, cookies, HTTP headers and bodies, URL query values, and database query data is disabled. Session replay is not enabled.
Analytics Data
We use the following analytics services:
- Google Analytics (GA4) — collects site usage data including pages visited, session duration, approximate geographic location, device type, and browser information. Public-site events can include the GPU model, count, region, and actions such as searches or alert interactions. Google Analytics uses cookies to distinguish browsers. You can opt out by using a browser extension such as Google Analytics Opt-out Browser Add-on. For more details, see Google's Privacy Policy.
- Vercel Analytics and Speed Insights — collect aggregate public-site usage and performance metrics. We exclude developer-portal routes from these services.
- PostHog — measures public-site page views, GPU searches, provider-link clicks, sharing, and subscription interactions. We also measure time spent on public pages, the maximum fraction of page content viewed, and numeric page-performance metrics, without collecting performance-entry details or element attribution. We use a random anonymous browser identifier in a first-party cookie (
__Host-gpufinder_experiment) for up to 90 days to keep experiment assignment consistent across visits and measure outcomes. Ordinary public-site events are not linked to your account or email. Clearing this cookie resets assignment. PostHog itself uses memory-only storage; we do not enable its cookies or local storage. The public-site tracker excludes developer-account pages and account identities. We include your browser's user-agent string to distinguish ordinary browser traffic from bots and automation. We record public-page sessions to understand navigation and usability problems, with input values and displayed subscription email addresses masked before transmission. File inputs, embedded frames, and developer pages are excluded from recordings. Replay page URLs omit query strings and fragments; console logs, network headers, and request and response bodies are not recorded. Recordings are retained for 30 days. Automatic analytics interaction capture remains disabled. Browser collection is disabled when Do Not Track or Global Privacy Control is enabled. We also count completed email confirmations for news and price alerts, using a new random identifier for each confirmation. These counts include the subscription type but no email address, confirmation token, browser identity, or visitor location. They are not linked to earlier site visits. For the optional search-watch experiment, we separately derive an experiment-only anonymous identifier from this cookie. We temporarily associate that identifier with a pending alert to count a confirmation once, then remove the association at confirmation or after 30 days. Experiment events contain the variant and outcome, never your email, account, subscription token, or ordinary browsing identifier. Experiment records are deleted after 90 days. This allows us to compare confirmed alerts and provider actions per exposed participant without joining confirmation to general browsing. Do Not Track, Global Privacy Control, and marked test traffic are excluded from this experiment. We also measure email delivery, delays, bounces, failures and spam complaints using Resend delivery receipts. These analytics contain message identifiers and fixed template/campaign labels, without email bodies, subjects, recipient addresses, or sign-in and unsubscribe links. Quota-notice receipts use the same hashed account identifier as API analytics; other email receipts use a separate identifier per message. We do not add email tracking pixels. On public pages, known email campaign labels can accompany page views and provider clicks for up to 30 minutes within the current browser document. Campaign labels do not identify an email recipient. Data is sent to our EU PostHog project. Separately, we send selected server-side account records to measure API activation and retention: account and key creation, successful API usage days, quota notices, extension requests, MCP authorizations, checkout creation, and billing outcomes. These use a hashed account identifier that is stable across visits and is not linked to anonymous public-site browsing. For signed-in developers viewing the API guide, we may test different onboarding entries and record the assigned version, visible entry, setup starts and reported setup difficulties against this same account identifier. We compare those events with successful API usage days to measure activation and return use. This does not connect your anonymous browsing identifier to your account. We exclude operator/test accounts and honor Do Not Track and Global Privacy Control for guide experiment assignment and events. Region, budget, credentials and other entered values are not included in these events. When you sign in or return with an active session, we also attach your verified account email to this PostHog account profile so we can recognize signed-in accounts. This profile update respects Do Not Track, Global Privacy Control, and account analytics exclusions. No API keys, sign-in tokens, or request contents are included in those account lifecycle events. We also sync Stripe customer, subscription, invoice, and payment records into PostHog for revenue analysis; those records can include billing contact details held by Stripe. The stable account identifier links API activity with billing. Contact us to request exclusion from account analytics.
MCP operational analytics are separate from account lifecycle events. When enabled, they send PostHog the internal account identifier for authenticated calls (or a temporary identifier for unauthenticated requests), request/session identifiers, client name and version, protocol and model information, tool names, call times, durations, outcomes and error details. They can also include tool arguments, a brief purpose supplied by the assistant, and tool responses, including their nested usage fields. We use these records to understand tool use and troubleshoot failures. The standard /mcp endpoint also returns a random conversation identifier that the assistant can send on later calls to group those calls into a conversation. This identifier is stored with MCP analytics. The /mcp/openai endpoint does not request or return conversation identifiers; it retains account-level and individual-call analytics. These settings apply to every caller of each endpoint. Do not include personal information or secrets in tool arguments or purpose text. You can contact us to request access to or deletion of account-linked analytics records. Revoking the connection stops future authorized calls but does not delete earlier records.
Developer-portal activity uses a separate, allowlisted event relay to GA4 in production. It sends the event type, outcome, portal area, a random browser client identifier, and a session timestamp; it does not send your email, account ID, API key, sign-in token, request body, or URL. This relay uses no browser cookies or storage and is disabled when Do Not Track or Global Privacy Control is enabled. The ordinary public-site page-view tracker excludes developer routes.
3. Cookies
Developer accounts use an HTTP-only, same-site session cookie to keep you signed in. The server session has a seven-day lifetime and may be refreshed while you use the portal. Google Analytics sets first-party cookies (such as _ga and _ga_*) to track browser sessions on our behalf. You can block these cookies using your browser settings or an analytics blocker without losing the core comparison features.
4. How We Use Your Data
- Email addresses are used for subscriptions, developer sign-in, and relevant API service notices; subscription preferences control the marketing or price alerts you choose.
- IP addresses and browser details associated with a developer session, and short-lived request records, are used to prevent spam and protect accounts.
- Provider-link activity is used to measure listing usefulness and maintain the service.
- Analytics data is used to understand how the site is used and to improve the service.
- Developer account and API records are used to provide access, enforce usage limits, investigate abuse, review extension requests, and reconcile billing.
- Stripe identifiers and payment status are used to provide and support paid plans.
- We do not sell, rent, or share your personal data with third parties for marketing purposes.
- We do not use your data for profiling.
5. Data Storage and Security
Subscription, request, developer-account, API-access and billing records are stored in a PostgreSQL database. Confirmation, unsubscribe and sign-in links contain security tokens, so you should not share those links. API key secrets are stored as hashes. Payment details are handled by Stripe rather than stored as card numbers in GPU Finder. We take reasonable technical and organizational measures to protect your data, but no method of transmission or storage is 100% secure.
6. Retention
- Unconfirmed subscriptions are removed after 14 days.
- Subscription IP addresses are removed after 30 days.
- Confirmed subscriptions remain active until you unsubscribe or ask us to delete them.
- Unsubscribed subscription records are removed after 90 days.
- Provider requests and outbound provider-link records are removed after 90 days.
- Temporary abuse-prevention counters are removed after they expire.
Developer account and API information is kept as needed to provide access, protect accounts, and reconcile billing. Revoking or expiring an API key stops its use but does not itself delete account, usage or audit records. Billing-linked deletion requests require operator review; backups and processor records may retain data after account records are removed.
Backups and processor logs may retain limited data for a short additional period before being overwritten.
7. Third-Party Services
GPU Finder uses the following third-party services that may process data:
- Google Sign-In — optional developer account authentication
- Google Analytics — website analytics (see section 2)
- PostHog — public-site, API activation, MCP operational, and revenue analytics (see section 2)
- Vercel — hosting, analytics, and performance monitoring
- Sentry — application error and performance monitoring
- Neon — managed PostgreSQL database hosting
- Resend — delivery of subscription, developer sign-in, and API usage emails
- Stripe — hosted Checkout and payment-method management when enabled for an account
Pricing data displayed on GPU Finder is aggregated from publicly available sources including the SkyPilot catalog and public cloud provider APIs. We do not send your subscription data to cloud providers. If you click a provider link, your browser contacts that provider directly and its own privacy policy applies.
8. Your Rights
You have the right to:
- Unsubscribe from alerts or updates at any time using the unsubscribe link in any email.
- Request deletion of your email address and any associated data by contacting us.
- Request access to the personal data we hold about you.
- Opt out of analytics by using browser privacy tools or extensions.
Developer accounts cannot be deleted through the portal. We handle developer-account deletion requests through support and review billing records before removal. Backups may retain data after account records are removed.
9. Children's Privacy
GPU Finder is not directed at children under 13. We do not knowingly collect personal information from children. If you believe a child has provided us with personal data, please contact us so we can delete it.
10. Changes to This Policy
We may update this Privacy Policy from time to time. Changes will be reflected on this page with an updated “Last updated” date. Continued use of the site after changes constitutes acceptance of the revised policy.
11. Contact
If you have questions about this Privacy Policy or want to exercise your data rights, contact us at hello@gpufinder.dev.