Privacy Policy

Updated 22 November 2026 · Provider commitments pending

Release draft. Privacy contact: Contact@xasri.com. Publishing configuration, release processing commitments and public-release reviews must be finalized before publication.

Who operates PRAXIS

Founders: PRAXIS is operated by its founders under the operating name Xasri Industries. The founders oversee its privacy, data protection and security practices, subject to applicable law. This does not represent the project as an incorporated company.

You can contact the founders about privacy, data protection, account-data requests or complaints at Contact@xasri.com. This monitored contact channel does not mean a statutory grievance officer, data protection officer or regional representative has been appointed. India and Google Play are the intended first release market and channel. Any legally required postal address, formal appointments and response procedures remain to be confirmed. This draft is not a declaration of legal compliance.

Information stored on your device

Regular tabs, browsing history, bookmarks, reading-list items, recent searches, custom shortcuts, preferences and download records are stored in the local browser profile. History recording and saved AI conversations are enabled by default and can be disabled in Privacy Center. History is capped at 500 entries, recent searches at 12 and saved download records at 200. Other saved records remain until removed; a general time-based retention period is not established.

Schedule Tasks stores the checks and reminders you request, their status, saved workflows, research collections, snippets and tab handoffs locally. Saved form profiles and drafts use platform secure storage. The password vault is also separate secure storage, with manual saving, showing and copying. Vault passwords and form values are not offered to AI tools.

Saved checks can revisit their chosen website while PRAXIS is open and active, so that website receives the resulting requests. Checks and reminders resume when you return; this build does not provide closed-browser monitoring or connected system notifications.

These local records support the features you choose. Browser history and bookmark synchronization is not implemented. Signing into XASRI does not synchronize this browser profile. People using the same operating-system/browser profile can access its ordinary saved data; use separate OS accounts for separate users.

Websites, cookies and permissions

Websites receive normal network requests, including your IP address, browser/runtime information, URLs, searches and any details you enter on the site. The WebView stores website cookies, caches and site storage. These belong to the visited websites and are distinct from XASRI account credentials and local browser records.

Supported site permissions ask before access and require a verified requesting origin and an active page. Grants are for the current session. Privacy Center lets you revoke recorded grants and reload affected pages to end access. Unsupported or unverifiable requests are denied. Unsolicited pop-ups are blocked; certificate checks remain enabled.

Clear browsing data separates history, recent searches, download records, local AI conversations, cookies, caches and supported site-storage clearing. Cookie clearing can sign you out of visited sites. A download-history operation leaves downloaded files on disk. Available website-storage scope is displayed before clearing; some platform capabilities remain limited.

Private browsing

Private tabs and their conversations are excluded from normal history, tab restoration and saved chats. Bookmarks and files you explicitly save can persist. Cloud AI in private tabs is off by default and requires a per-tab decision when enabled. Sending a private request to an external service is still external processing.

Windows uses WebView2 InPrivate storage. Mobile private-cookie isolation depends on the platform capability and must be verified before release; this draft does not promise that mobile private tabs have independent cookie stores. Private browsing does not hide traffic from websites, network providers or administrators.

AI requests and external processing

When you use the cloud AI assistant, your submitted prompt goes to XASRI Cloud at api.xasri.com. Before the first assistant request under the current notice, PRAXIS explains the processing and asks for an affirmative choice. Your approved notice version is recorded locally; completing the general introduction alone does not approve assistant processing. Optional controls govern sharing relevant page text and URL, selected text, explicitly selected tabs, and uploaded documents. Context sharing is off by default and is requested when you choose the relevant action. Disabling page sharing preserves ordinary browsing and prompt-only chat. XASRI search has a separate request path described below.

XASRI Cloud uses the enabled Sarvam AI service for supported chat, research and coding, and for requested document reading and synthesized speech. Submitted prompts, relevant conversation context and the page, text or file content you choose to share may be processed by the applicable provider. Requested search and research can send queries and requested URLs to available services such as Tavily, Exa and Brave Search, and to public reference sources such as Wikipedia, arXiv, Crossref, OpenAlex and NASA. A configured crawling service may receive a requested website URL. Provider availability and routing depend on the service; this does not mean every provider receives every request. A separately configured optional backend can use OpenAI, Anthropic, Google Gemini or Groq. Those optional integrations are not a statement that they are enabled in the current XASRI Cloud release.

Signed-in XASRI features can derive account-linked facts, preferences or goals from submitted prompts to support later answers. These cloud account memories are separate from the browser’s local remembered-chat setting. Supported XASRI services also store account profiles, saved conversations, messages, research task records and attachments. These account records have no general automatic time expiry in the reviewed implementation and remain until the relevant deletion operation removes them. Not every browser request creates a saved cloud conversation. Account deletion includes associated application data and account memories; provider-held copies have a separate scope.

Provider terms can permit use of submitted content or queries for service improvement or model training. Sarvam’s published privacy policy describes an account-setting or written training opt-out and a configurable content-retention default of 30 days after last access. Its account and technical records have different periods. Tavily describes query improvement unless its customer contract says otherwise and retains information according to continuing service needs or valid deletion requests. Exa describes query training and separately governed business-customer processing. These published terms are not proof of the settings or agreement applied to XASRI’s accounts. A no-training agreement, opt-out, processing location or downstream deletion guarantee for those accounts has not been verified. This review therefore does not promise zero retention, training exclusion or processing entirely on your device. Provider and infrastructure records may be processed across borders. Provider documentation is linked below; the applicable account settings and commitments must be confirmed before public release.

Document reading asks before each selected file is uploaded. Supported PDF documents and selected PNG, JPG, JPEG or WEBP images can be sent through XASRI to Sarvam for processing. Extracted text may be included in a later AI request. The reviewed browser flow does not establish deletion of the provider’s processing job or every provider-held copy after a response. Avoid submitting credentials, verification codes or unrelated private information. The app excludes sensitive form fields at their source and applies secondary redaction, but it cannot reliably recognize every secret pasted into free text or a document.

Webpages, search results and files are treated as untrusted context. They cannot authorize tools or confirm actions. Agent actions pass a shared permission gateway. Personal form filling follows local inspection, detail collection, review, permission and value verification; submission stays manual. Sensitive password, OTP, CAPTCHA, payment-code and identity-verification fields stay manual.

Search, icons, voice and downloads

The regular homepage shows public BBC India news headlines in Discover by default. The feed and thumbnails are fetched from BBC hosts. These hosts receive ordinary IP/network metadata, not your browsing history, typed searches, account tokens or website cookies. Articles show the publisher’s publication time and open the original BBC link when selected. Discover can be disabled from its menu. Incognito and child mode do not load this feed. This setting is separate from typed-query suggestions and remote website icons.

Ordinary searches go to your selected search engine. XASRI search sends your query to XASRI Cloud and remains available separately from the cloud AI preference. Optional typed-query suggestions send the text to Google and may fall back to DuckDuckGo. Optional remote website icons send a domain to the site or Google favicon service. Both suggestions and remote icons are off by default and can be controlled independently. Cached icons can remain locally until the cache is cleared. Requested page translation first explains that Google Translate will receive the selected page URL and may fetch its public content. After your approval, PRAXIS opens that translation URL; it does not send the website’s cookies or form values. These controls do not prevent requests needed to visit a website or use an enabled search service.

On Android, voice input uses only the on-device speech recognizer on Android 12 or later with an available local model. It does not fall back to external recognition; voice input is unavailable when that local recognizer is unavailable. Android Device read-aloud selects an installed voice that does not require a network connection and stops if such a voice cannot be selected. The recognized text is sent to XASRI only when used for a requested cloud feature. Optional Sarvam speech synthesis sends answer text, language and voice settings over HTTPS through XASRI and returns generated audio; it is not a recording upload. Temporary app-generated playback files are removed after playback, stop or failure. Other platforms use their platform speech services, whose external processing depends on that service. Microphone access requires OS permission.

Downloads contact the selected source and may use that website session for authorized files. Download records and files have separate deletion controls. Risk and file-type checks reduce accidental opening of dangerous content; no malware or reputation scanner is configured, and PRAXIS does not guarantee a file is safe. Private files are not uploaded for scanning.

Accounts, diagnostics and optional services

XASRI authentication processes your sign-in and profile information, with account authentication and supported data storage provided through Supabase. OAuth access/refresh credentials use platform secure storage. Account and guest AI-use counters support service access and quotas; a random guest device label can accompany cloud, report and allowance requests, including the allowance refresh when you open the account page. Guest lifetime-budget records do not have an automatic expiry or a verified guest server-deletion control in the reviewed service. The service’s request audit can contain an account identifier, request identifier, time, method, route, status and duration. Hosting, database and platform services have separate logs and backups whose retention has not been established by the local browser controls.

Optional first-party search measurement is off by default and requires a configured service and your opt-in. It sends activity dates, explicit browser search dispatch counts, the chosen engine, Android or Windows, app version and a country derived only from your optional region setting; unspecified regions remain unknown. A random installation identifier, separate from your account, supports deduplication, active-install counts and deletion. No search text, page URL, browsing history, private-tab or Family browsing events are included. The receiving server sees ordinary network metadata. Counts are browser-reported and do not represent Google-verified traffic or revenue.

Measurement waits locally for up to 7 days and 500 events during connection failures. Turning sharing off clears pending local events and stops collection. Delete shared measurement data also requests erasure of the installation and events from the configured service; the interface reports whether deletion was confirmed. Operational backups and previously exported reports require separate operator handling. The private reporting service applies its configured bounded event retention, documented for the deployment.

Local agent safety records contain only action/tool category, permission decision, confirmation status, outcome and timestamp. They exclude prompts, page contents, arguments, URLs and form values. Retention is 7 days by default, with 0, 7 or 30 days selectable, capped at 500 events and pruned when the app accesses the records. Closing the app is not immediate deletion of all local records.

Report AI content opens an in-app form. It sends the reason, optional details and only the answer excerpt you explicitly choose and can edit. A reason-only report is supported. Signed-in reports use your account session; guest reports use a random device budget label. The app confirms receipt only after the reporting service returns a valid receipt. Application-managed reports and moderation notes expire after 90 days, and hash-only report abuse receipts after 24 hours. The reporting service and private persistent store have passed a synthetic receipt check; that check does not promise immediate review or removal of reported content. No report is claimed received when the service is unavailable.

The optional backend uses authenticated client sessions and owner-scoped workspaces. Its process-local safety records are capped at 500 events and 24 hours. Workspace files are ephemeral and containers have a two-hour lifetime; application-managed closure attempts earlier cleanup. Deployment logs, backups and cloud service retention remain the operator’s responsibility and need release verification.

No payment or subscription checkout is implemented in this browser. An account plan may be fetched for display. Billing and transactions performed on third-party websites are governed by those services.

Retention, deletion and your choices

Privacy Center lets you disable optional AI processing and sharing, change saved-history/chat preferences, clear selected local categories, manage saved form profiles and drafts, revoke supported permissions, view/clear safety records, and export supported regular browser records. Turning off remembered chats removes saved local conversations. Export excludes credentials, cookie contents, private tabs and secure form values.

Signing out removes this app’s account session and attempts server token revocation. It does not close the XASRI account, delete website accounts or erase ordinary browser history. Account settings separately let you initiate XASRI account deletion after recent sign-in and explicit confirmation. The enabled service deletes the authenticated account and associated active application data across XASRI services; it reports success only after confirmed cleanup. The public request resource is https://api.xasri.com/v1/praxis/account-deletion and also explains how to request help without the app. A disposable-account test confirmed removal of its account, tested associated rows and reports, and denial of its old token. This test did not cover every uploaded object, provider copy or backup. Unavailable or incomplete requests are not reported as deleted. Contact@xasri.com remains the contact for requests and help.

The account-deletion implementation retains an irreversible SHA-256 account digest with deletion status and time indefinitely. Hash-only task deletion markers are also retained indefinitely to prevent late work from recreating deleted records. Updated service processes block account access; their private process memory is physically cleared on restart. Provider and infrastructure logs, backups, legal retention and another person’s exported copies have separate retention and handling. Local browser records and downloaded files remain under separate Privacy Center controls. Immediate universal erasure or zero retention is not promised; live operator and provider commitments still require verification.

Rights, regions and complaints

Depending on where the product is offered and which law applies, you may have rights to access, correction, deletion, withdrawal of consent, information about processing, portability, objection or restriction, and complaints to a competent authority. Local controls are available regardless of region. Send privacy and rights requests to Contact@xasri.com. Additional request workflows must be established before launch in a jurisdiction that requires them.

India’s DPDP framework has phased commencement; EU/UK data-protection rules and California requirements apply when their legal scope is met. Selecting a broad country/region helps show relevant rights information and legal viewing options. The app does not use precise location to make that selection or treat every user as subject to every law. Applicable markets, legal bases, transfer arrangements and required response procedures remain subject to review.

Contact@xasri.com is the monitored privacy and complaint contact confirmed by the founders. Request verification, response ownership, legally required response times and escalation procedures still need to be documented and verified before launch. No unverified postal address or formal officer appointment is published here.

Children and protected users

PRAXIS opens regular browsing by default without an age-group or date-of-birth question. It does not collect age information to select a browsing mode; an unspecified local profile allows regular browsing and does not identify the user as an adult. Family browsing is an optional setting on supported Android devices. Existing saved child profiles and active Family locks stay protected until parent authentication. The child process uses separate WebView storage and does not start XASRI accounts, guest AI identifiers, cloud AI/OCR, agents, voice, Discover, remote suggestions or remote icons.

Child browsing is limited to parent-approved HTTPS reading sites and approved resource hosts. Scripts, forms, frames, pop-ups, uploads, microphone and location are disabled. Requests and redirects pass local destination controls. Browsing history is not recorded; parents can clear child website data. Websites still receive ordinary traffic and may change their content.

Parents use the Android device credential for protected actions and leaving child mode. An authenticated parent can return to regular browsing without providing age information. A parent may explicitly hand a download to another app, outside child controls. Device authentication is not legal verifiable parental consent. Applicable Google Play Families review, child-data consent, provider eligibility and rights processes remain release requirements. The app does not claim verified age assurance or certified child safety.

Ad filtering and rule updates

Android and Windows webpage filtering use the Brave engine with local EasyList/EasyPrivacy rules and exact-host site exceptions. Matching and supported cosmetic filtering happen locally. Windows private tabs ignore ordinary-profile site exceptions. Pop-up permissions are separate from ad-filter exceptions. Filtering applies inside PRAXIS, not to other applications. No PRAXIS ad SDK, paid allowlisting, replacement ads or affiliate-link injection is included.

The application bundles initial rules. Windows rules update with application releases. Android signed network updates require a configured publishing endpoint and public key, are bounded and verified, and retain the previous rules if an update fails. Update requests do not include browsing URLs, cookies, referrers, account tokens or installation identifiers; the host can receive ordinary IP/network metadata. Child mode reuses local rules and makes no update request. Rule publishing and server retention must be verified before enabling production updates.

Security and policy updates

Account, cloud AI, requested search, file upload, content-report and cloud speech requests require HTTPS, and the app does not follow API redirects. App-rendered remote images and server-side AI website retrieval also require HTTPS and reject plaintext redirects. XASRI internal service traffic uses its hosting platform’s encrypted private network. Ordinary open-web navigation follows the visited website’s connection and is separate from these app-controlled data transfers.

The implementation uses secure credential storage, authenticated backend operations, scoped tool authorization, reviewed consequential actions, source exclusion/redaction, network-destination checks for backend research, and bounded safety logs. Existing OS/WebView security protections remain enabled. These safeguards have limitations and cannot guarantee perfect security or legal compliance.

This draft’s version and update date appear at the top. Material processing changes require updated information and relevant consent before the new optional transfer. The release operator must finalize policy publication, incident handling, contracts and legally required notices. Government legal notifications are internal verification sources; they do not create user alerts or notifications in this browser.

Third-party privacy documentation

Recipients depend on the feature and enabled service. Not every provider receives every request.