Privacy Policy

FidelicAI does not use cookies, run Google Analytics or Meta Pixel, sell your data, or build behavioral profiles. Cookieless traffic measurement and information you choose to submit are the limited exceptions, with each data path named in plain language.

Who we are

FidelicAI is a DBA of Memetica Corp, a corporation incorporated in the state of New York. We operate this website (fidelic.ai). Questions about this policy go to hello@fidelic.ai.

Cookies and tracking

We do not set cookies on this site. We do not load Google Analytics, Google Tag Manager, Facebook Pixel, LinkedIn Insight Tag, or any other behavioral-advertising pixel. There is no consent banner because there is nothing to consent to.

Two cookieless measurement tools run on the site so we can see traffic volume and page performance:

• Vercel Web Analytics: counts page views and shows top referrers and countries. It does not use cookies and does not collect personal identifiers. See Vercel's analytics privacy notice.
• Vercel Speed Insights: measures Core Web Vitals (page load, input delay) anonymously. Also cookieless.

Microsoft Clarity is wired in but disabled. Session replay will not run until this policy names the recordings, fields, purpose, and retention period beforecollection begins.

What we collect when you actively give it

We do collect data when you take an action that requires it. Specifically:

• Email updates and waitlist signups. When you submit the email form in our footer or on a waitlist page, we store your email address and the context around the signup: which page you signed up from, which source brought you here (UTM parameters), the referring site, and your first landing path. This lets us understand which essays and pages actually move people to subscribe. We use this for our own analytics and to send you the updates you asked for. We do not share this list.
• Cookbook requests. We store your requested business problem, email address, source page and referral context as an authenticated-only Sanity draft. We use it for publishing research and updates about your request. The request is not published in our catalog or AI-readable files and does not join a general mailing list. Keep customer records, passwords and other private business data out of the request.
• Website workshop requests. We store your email, public website address, source and referral context, and the result of an automated public-site check as an authenticated-only Sanity draft. The check reads public pages and sitemaps to assess whether the site suits the two-hour workshop. Suitable sites receive calendar access immediately. If your site needs a broader session, we keep the request to follow up when an appropriate workshop is available. Name, coding-agent details, format, times and scope acknowledgment are stored only when supplied, including in older request forms or later booking steps. We use a keyed visitor digest to limit repeated checks, without storing your raw IP address in those counters. Your private calendar-access token is kept in this tab’s session storage so a refresh does not lose access. A request does not join a general mailing list, reserve a time or charge a card. Payment details belong in Stripe after the time and booking terms are agreed. Keep credentials and customer records out of the request. You can ask us to delete your request using the deletion contact in this policy.
• Workshop reservations and agreement records. We keep the agreed time, format, office location or video link, signer and business name, accepted agreement text and version, acceptance time, and payment reference IDs in authenticated-only operational records. Stripe stores card details; the website stores provider reference IDs and payment status, not card numbers. We use these records to reserve the session, administer authorizations, agreed fees and refunds, and retain evidence of the agreement. Google receives the private host reservation time and session details in a separate workshop calendar. We do not add the client as a Google attendee or send Google invitations. The client can download a separate calendar addition without the host’s account email. Booking links are private access links; keep them confidential. Agreement and payment records may need to be retained for legal and accounting obligations after other request details are deleted.
• Work-order requests. When you submit a work request, we store the work email, work category, timing, source page, optional note, and the same referral context: source page, UTM parameters, referring site, and first landing path. We use the record to assess the scope and reply. We do not add it to a general mailing list. The form is not a file upload: do not include protected health information, patient data, provider identifiers, résumés, personnel records, contracts, payment details, passwords, authentication codes, identity documents, tax records, portal or account credentials, customer data, private analytics exports, proprietary reports or files, access invitations, or other confidential material in the note.
• Emails you send us. Anything you write to hello@fidelic.ai is read by people who work at FidelicAI and stored in our inbox for as long as the conversation is useful to continue.
• Server logs. Our hosting provider (Vercel) records standard request logs: IP address, user agent, requested path, timestamp: for short retention. These are operational logs used for debugging and abuse prevention. Access to the website is restricted by estimated IP location to the contiguous United States, including Washington, DC. Verified search and AI crawlers and authenticated service endpoints have specific exceptions. Blocked-request logs include the IP address, estimated country and state, requested path, host, method, user agent, timestamp, request identifier, and blocking reason. We use them to analyze access restrictions and unwanted traffic. These application logs exclude query strings, request bodies, cookies, and authentication headers. Vercel Pro retains runtime logs for one day; we do not maintain a separate blocked-IP archive. Vercel describes its data handling in its privacy policy.

Older discovery connector

The existing discovery connector at mcp.fidelic.ai lets Claude conversations find public FidelicAI AI agents, open an accurate hire path, read FidelicAI publishing, and reach official learning, support, and community routes. See the connector page. Its data path is separate from the expertise beta. The existing discovery connector has its own scope and data practices. Its data path is narrower than the website's:

• No accounts, no sign-in, no cookies. The connector has no user accounts and sets nothing in your browser or your Claude client.
• What a request contains. Claude sends only the bounded fields for the selected action. Those fields can include a roster category or search phrase, a public agent id, an engagement option, a publication type or search phrase, a public fidelic.ai page address, a maximum reading length, or optional non-secret setup answers for a supported hire. Setup answers can include a show name, public show link, expected volume, preferred Slack channel name, recording-tool name, first assignment, target date, Drive deferral, and recording readiness. The connector does not ask for your name, email address, customer workspace identity, payment data, or credentials. We do not store request bodies or tool arguments. Do not put confidential business information in a search or setup field.
• Prepared setup links. When you ask Claude to prepare supported setup answers, the connector encrypts those bounded fields into a link that expires after 30 minutes. The encrypted token can appear in the Claude conversation, browser history, and ordinary infrastructure request logs, so it is deliberately limited to non-secret setup information. start.fidelic.ai decrypts it on the server, removes it from the browser address after the page loads, and shows the answers for review. The token cannot record an agreement, authorize Slack or Drive, approve payment, or approve deployment. After sign-in, prepared answers and the `claude-mcp` acquisition source enter the saved onboarding draft and are shown for review; identity comes from the verified account.
• Operational logs and retention. The application does not log request bodies or tool arguments. Current site and workshop hosting uses Vercel. Vercel processes ordinary HTTP and application metadata, which can include source IP address, user agent, host and path, method, timestamp, response status, timing, response size, and a bounded application error message. Vercel retention follows the active plan and settings; see its runtime log guide. Vercel's firewall also processes network metadata for abuse prevention. The application temporarily counts source IP addresses in each running instance; these counters clear on idle eviction or process restart. On Vercel, an additional firewall limit counts requests by source IP over 60-second windows in each region. These are regional counters, not one worldwide total, and no additional IP database is added to the application. Railway documents its log fields and plan retention in its logging guide.
• First-party public page reads. Role and publication actions can retrieve a public page from fidelic.ai. The connector accepts only the configured fidelic.ai origin, refuses redirects, and limits time and response size. The search phrase itself is not forwarded to the website. Vercel processes the ordinary site request because it hosts fidelic.ai.
• Content-free operator notices. After a client successfully starts the connector or asks for a valid hire path, the service can send FidelicAI an internal notice through Resend. The notice contains only a fixed event label and server timestamp. It does not include the source IP address, user agent, client name, request body, search text, setup answers, or other submitted fields. A temporary one-way source digest suppresses repeat notices and disappears when the process restarts.
• Nothing is sold or used for advertising. Railway and Vercel process the limited operational data above as infrastructure providers. Resend can process the fixed operator-notice label and timestamp as our email provider. We do not sell request content, send it to an advertising service, or use it to build a behavioral profile. A hire finishes at start.fidelic.ai, where the data practices for account, checkout, and workspace connection apply before you submit them.
• Community membership is separate. The connector can return the stable FidelicAI community link. It does not call a Slack API, add you to the community, connect a customer workspace, or share customer channels, files, messages, or agent work.

Website cookbook connection

No business data is required to read cookbook recipes. Your AI sends a short search or recipe identifier; the service processes it in memory to return guidance. The application does not store request bodies, tool arguments, conversation histories, customer projects or session records. Keep files, credentials and private evidence in your own AI environment.

The connection does not give FidelicAI access to your files, website or business accounts. Railway is no longer retained as a rollback host. Hosting providers process operational metadata as described above and in the cookbook privacy notice. Your AI provider has its own data practices.

The cookbook is free during testing. Optional email updates and cookbook requests use the separate website form handling described above; neither is required to read recipes. Future paid plans will have their own terms before checkout opens. See the connection guide and AI querying instructions.

Where your data lives

• Sanity stores our website content, waitlist signups, cookbook requests, and work-order requests, including their source and referral context. Hosted in the United States. See Sanity's privacy policy.
• Vercel hosts the site and runs cookieless Web Analytics and Speed Insights. Its operational logs and regional firewall counters support debugging and abuse prevention. Workshop payment handlers do not log card details or acceptance text.
• Resend / Postmark deliver any transactional or update email we send you. Resend can also deliver the connector's content-free operator notices described above.
• Former hosting. Railway previously hosted services. Those projects are retired and are not part of the workshop setup. This statement does not confirm deletion or recovery of historical provider logs.

We do not use third-party advertising networks. We have no data brokers in our stack.

AI engines and crawlers

FidelicAI is content that wants to be read: by people, and by AI engines that summarize and cite. We allow GPTBot, ClaudeBot, PerplexityBot, and Google-Extended to crawl us. Our llms.txt explains how the site is organized and what to cite. Our robots.txt lists which crawlers we welcome. None of this involves your data: only ours.

Your choices

• Remove yourself from the email list. Every email we send has an unsubscribe link. You can also email hello@fidelic.ai and we will remove you and delete the record.
• Request your data or its deletion. Email hello@fidelic.ai. We aim to respond within seven days. We will ask you to confirm the email address on file so we can match the record.
• Block analytics in your browser. Vercel Web Analytics respects standard tracking-prevention signals; if you use a privacy-focused browser or extension, the analytics endpoint will not load. The site itself will work exactly the same.

Children

This site is not directed to people under 16. We do not knowingly collect data from children.

Changes to this policy

We will update this page when our practices change. The "last updated" timestamp at the top reflects the most recent revision. Material changes will be announced in our email updates before they take effect.

Related: Terms of Service · Security.