Privacy Policy

1. Who we are and what this policy covers

“UrAI,” “we,” “us” and “our” refer to the independent operation of the UrAI website and chat app. As the service operator, we're responsible for the processing we control. Our privacy contact is support@urzen.site.

This policy applies to visitors, guests, signed-in users, people mentioned in submitted content, viewers of shared links and anyone who emails support—both on the web and in the installed PWA. External sign-in providers, AI labs, advertisers and third-party sites have their own policies. Features described here may vary by model, region or account, and our Terms of Service govern how you use the app.

2. Key points before you use UrAI

• Signed-in chats are saved to your account history. Guest chats aren't saved as account history, though requests still go through our server and the chosen AI provider.
• Private account chats can still be stored temporarily on the server. Technical logs and provider retention still apply—private mode isn't end-to-end encrypted.
• Relevant chat history, custom instructions, memory, uploads and search results may be sent with a request. Extra background calls (like generating a title, planning a web search or summarizing memory) can happen automatically.
• Sharing a link publishes a snapshot anyone with the URL can view and copy. Revoking the link won't delete copies others have already saved.
• Browser IDs, fingerprint hashes, IPs, cookies and usage counters keep the service running, secure and within limits. Sharing an IP with someone else doesn't merge your guest allowances.
• The main app loads Google AdSense, which handles device, network and cookie data separately from the chat backend.
• Don't submit passwords, API keys, sensitive records or someone else's private info unless you're authorized to share them.

3. Information we process

Account and authentication information

Signing in with Google gives us your Google ID, email, name and avatar URL. GitHub sign-in gives us your account ID, username or display name, avatar and public email (depending on your settings). We never see your Google or GitHub password.

We store an internal user ID, profile settings and any display name or avatar you pick. Login sessions store a protected token representation, timestamps, expiry, user-agent and the IP address used when the session started.

Content and preferences

We process whatever you send or generate: messages, replies, titles, edits, code, file names, attachments, images, masks, video, voice transcripts, speech requests and custom instructions—along with model names, source links, cards and provider reasoning summaries. Preferences (language, tone, model choice, UI layout, “about you” notes, memory toggles and voice settings) are saved to your account or browser depending on the setting.

Guest, device and usage information

Guest mode uses a server-issued guest ID, a signed cookie and a browser ID. The app also runs the bundled FingerprintJS library locally to build a browser fingerprint from technical traits (browser capabilities, platform, screen, language, graphics, audio, fonts), which is sent to UrAI and stored as a hash. Optional library telemetry is off, and we don't use Fingerprint's hosted API.

We also log operational data: IP addresses, user-agent, URLs, latency, status codes, model/provider picks, request counts, estimated usage, live-session duration and errors. Guest IPs and fingerprints are generally stored as hashes (which are pseudonymous, not anonymous), though hosting, security and session layers can still see raw IPs.

Support and reports

If you email support or report an issue, we keep your email, message and attachments to handle it. Technical logs aren't meant to be a chat archive, though URLs or error traces can occasionally include query fragments.

4. Guests, signed-in history and private mode

Guests

Guests can chat without signing in. Messages stay in your browser session and are sent with context when you ask for a reply, without being saved as account history. Refreshing, closing the tab or clearing site data can wipe an unsaved thread. Guest IDs, usage records and provider requests still apply.

Registered account conversations

Normal signed-in chats and attachments are saved in our database and storage so you can access them across devices, edit or delete them, and use memory. Relevant earlier messages are sent to the model for context, up to the active context limit.

Private mode

Private mode hides a chat from your normal history list and changes how memory works. For signed-in text chats, the server still temporarily stores the private chat, its messages and attachments. They become eligible for cleanup after two days of inactivity, and cleanup runs during normal service operations rather than on an exact timer.

Leaving private mode clears the local view, not necessarily the server record immediately. Private chats can't be shared via public links and don't feed into the cross-chat memory index, though existing preferences/memory can still be read if you turn on private personalization. Usage tracking, security logs, AdSense on the page and upstream provider rules still apply.

5. AI processing, additional requests and model training

To answer you, we forward relevant Input and context—current message, earlier turns, system prompt, custom instructions, retrieved memories, uploads, images, video and search snippets—to the chosen provider or router. Background helpers (deciding whether to search, interpreting image prompts, titling chats, summarizing memory) can trigger extra model calls before a reply appears. Retries may re-send the request, and a model's display name doesn't guarantee a specific physical server location.

Upstream providers enforce their own safety, fraud, retention and data-use rules, which vary between free and paid tiers, regions and endpoints. Routers like OpenRouter pass requests on to underlying inference hosts. UrAI can't guarantee zero retention, no human review or no model training across every third-party route.

UrAI doesn't train its own foundation models, and we don't sell or publish your private chats—but that doesn't override the terms of the AI provider handling your request. Check section 10 before sending anything that requires strict confidentiality or data residency. Displayed reasoning text is provided by the model/feature and may be saved with the reply, not as a complete view into how the model works internally.

6. Cross-chat memory and personalization

For signed-in users, memory can pull facts and short excerpts from normal saved chats to personalize later replies (like your interests or preferences). Entries and summaries may be generated by an external AI model and can occasionally be incomplete or inaccurate.

In settings, you can view, disable or clear memory, or forget specific items. Clearing memory stops old chats from being re-indexed, without deleting the chats themselves. Deleting a saved chat removes its linked message-memory rows, though facts repeated in other chats or existing summaries may still need manual clearing—or a support request if you want a full wipe.

Guests don't get cross-chat memory, and private chats don't add to it (though you can opt in to reading existing memory inside a private chat).

7. Files, images, microphone, camera and screen sharing

Documents and uploaded media

Uploaded files can be parsed, converted or run through OCR on our server before sending extracted text or media to a model; supported video can go to Google's native media endpoint. Attachments in saved chats live in Cloudflare R2 or protected local storage and are fetched via short-lived signed URLs (anyone with an unexpired signed URL can open it until it expires; some legacy media works differently).

Generated images and avatars

Image generation and editing send your prompt, reference image, mask and size to Cloudflare's AI service. Outputs and avatars are stored in R2 or local media storage. Generated images and avatars may be served over public media URLs, so anyone who gets the URL can view the file—even if you generated it in a private chat.

Dictation and speech playback

Voice dictation goes to Groq's speech-to-text API (with a local fallback where configured) to return text; the transcription step isn't meant as a permanent audio archive. Text-to-speech playback routes through OpenRouter's speech endpoint.

Live conversations

Live voice streams audio (and camera or screen frames if you allow them) through UrAI to Google's Gemini Live service, along with relevant chat context. Make sure anyone else visible or audible gave permission before you start capture.

In normal account chats, completed voice turns can be saved as text. Private live sessions skip normal transcript saving, though technical connection records (encrypted session instructions, encrypted reconnection tokens, timestamps, model, duration and usage) are still kept briefly. Stopping capture won't recall audio already sent upstream.

8. Public conversation links

Sharing creates a public snapshot anyone with the URL can view without an account, including messages, generated images, sources, cards and attachment metadata or URLs. Check the thread carefully before sharing.

Snapshots don't auto-update when you edit the original chat later. Other users can import a copy into their own accounts or save and share it elsewhere. Revoking a link stops access through that URL, but it can't delete copies people have already saved. Private chats can't be shared.

9. Why we process information and applicable legal bases

Here is why we process data and, where European or UK privacy law applies, the legal bases we rely on:

Without the core data needed to run a feature (like a prompt or a login signal), that feature simply can't work. Optional stuff—profile notes, uploads, memory, sharing, voice/video—is up to you. Don't put regulated sensitive records (medical, etc.) into general chat.

10. Who receives information

We only send data to the providers needed for the feature you use. Remember that a model's brand name doesn't always mean we connect directly to that lab—a router or cloud host may handle the request.

We don't dump your whole database to providers, though search queries, retrieved memory or uploads can carry more context than a single prompt. External images, avatars, search links, ads and previews can also make your browser contact third-party hosts directly. We may disclose data when legally required (e.g., valid legal process, abuse investigations or service transfers), and we don't pass private chat text to advertisers.

11. Where information is processed

UrAI's app server is hosted in the United States alongside a hosted database, while Cloudflare and our AI, auth, search and ad providers operate across various countries. Picking a specific model or using a VPN doesn't guarantee a particular processing country or data residency.

Where legal transfer rules apply, transfers rely on recognized mechanisms like adequacy decisions or standard contractual safeguards—though as an independent project, we don't claim custom enterprise DPAs or formal transfer certifications on every route. If you need details about a specific provider or safeguard before sending data, email support@urzen.site.

12. Cookies, local storage and application caching

Cookies are small values stored by your browser and sent with requests; local storage keeps app state on your device; a browser fingerprint is a derived technical ID, not a cookie. Clearing one doesn't automatically clear the others.

You can inspect or wipe cookies, local storage, caches and permissions in your browser settings. Blocking essential storage may break sign-in, preferences or guest continuity, and clearing browser data won't erase server-side account records or data held by third-party providers. Where optional cookies or tracking require consent under local law, we rely on actual consent rather than treating a notice or continued browsing as permission.

13. Google AdSense and advertising choices

The main UrAI app loads Google AdSense, so your browser may contact Google as soon as the app opens—sending network info, browser details and the page URL, and letting Google and its partners use cookies or ad identifiers where allowed. These standalone legal pages don't load AdSense.

Third-party vendors, including Google, may use advertising cookies to serve ads based on your visits to UrAI and other sites, as well as for measurement, frequency capping and fraud prevention. See Google's business data responsibility page, how Google uses partner-site information and Google's advertising info for their side of the processing.

Manage advertising preferences

• Use Google My Ad Center and Google Ads Settings to manage ad personalization.
• Check Digital Advertising Alliance opt-out tools or individual vendor controls where supported in your browser and region.
• Where a consent banner is shown, use it to accept, reject or update your choices.
• You can also block third-party cookies or ad scripts in your browser (opting out of personalization may still show contextual ads and won't delete data a vendor already holds).

In regions like the EEA, UK and Switzerland, optional ad storage and personalized ads require valid consent; if a consent control looks missing or broken, email support. Some regional privacy laws treat ad-related data sharing as a “sale,” “sharing” or targeted advertising—where those apply, use the opt-out controls above or contact us, though we don't currently claim automatic support for every browser Do Not Track or Global Privacy Control signal.

14. How long information remains

Retention depends on what the data is. Note that cleanup jobs often run during app activity or startup rather than deleting records to the exact second. Here's how our production setup handles retention:

If a legal requirement forces us to keep something longer, we'll let you know when handling your request.

15. Your controls and deletion requests

In the app, you can delete chats and messages, clear history, revoke share links, edit your profile/personalization and manage or clear memory. Keep in mind those controls do different things—clearing memory doesn't wipe old chats, deleting a saved chat doesn't pull back public media URLs or provider copies, and signing out doesn't delete your account.

There's no one-click full account-deletion or full data-export button yet. To delete your account or ask for a full export or correction, email support@urzen.site. If you're a guest, email us before clearing your browser profile so we can help locate your records—and please don't send us your raw signed cookie value.

16. Privacy rights and how to exercise them

Depending on where you live, you may have rights to access, correct, delete, restrict or port your data, withdraw consent, object to processing (including legitimate-interest processing or direct marketing), opt out of targeted ad sharing, or file a complaint with a regulator.

To make a request, email support@urzen.site with “Privacy request” in the subject, your account email or a non-secret ID, and what you want done. Don't send passwords, full cookies or ID documents unless we've specifically agreed on a safe way to verify ownership. Under European rules, the standard response window is one month (extendable for complex requests), and we won't penalize you for exercising your rights.

You can also complain directly to your local data-protection authority—see the European Data Protection Board list for EEA authorities or the UK ICO.

17. Security and limits of confidentiality

We use HTTPS, account access checks, signed guest cookies, protected session tokens, HttpOnly/Secure auth cookies in production, signed attachment URLs and encryption for stored live-session secrets. Still, messages aren't end-to-end encrypted—our server and upstream AI providers have to read your prompts to answer them, and no online service can promise zero risk.

Public image URLs, shared links, browser extensions and unsafe devices can expose data too. Lock down your Google/GitHub account, check files before uploading and email support if you spot a security or privacy issue.

18. Children and other people's information

UrAI is for users aged 18 and over, and we don't knowingly collect data from children. If you think a minor is using UrAI, email support so we can look into it and remove the data. Also, only upload someone else's voice, photo or personal info if you have permission—and if your info appears in someone else's shared link, contact us to get it reviewed.

19. Automated processing and decisions

AI answers, search routing, memory extraction, chat titles, safety filters and usage limits are all automated and can occasionally get things wrong—email support if a limit or filter misfires. UrAI isn't built for hiring, lending, housing, healthcare or other legally significant decisions, so don't rely on it for those.

20. External links, embedded resources and generated previews

Search citations, company logos, remote avatars, ads and generated HTML/document previews can trigger requests to outside hosts, exposing your IP and browser info to those sites. Don't type passwords or personal info into AI-generated previews, and check their code before trusting them.

21. Updates to this policy

We'll update this policy when features, providers or laws change, and bump the date at the top. For major changes, we'll post a notice in the app (and ask for fresh consent if required). If you're viewing an offline PWA copy, check ai.urzen.site/privacy for the latest version.

22. Contact UrAI

Use “Privacy request” for data-rights requests and “Security report” for security issues, and include only the non-secret details needed to find your account or link.