Privacy policy

What stays in your account, what AI receives, and the choices you have.

Updated 9 October 2026 · Version 2026-10-09.6
All legal documents

Email activation and two-step sign-in

New registration requires an email address and a six-digit activation code before an account is created. Account activation alone does not grant a new trial; eligible trials start after completing a plan subscription in Stripe. Existing accounts can keep signing in and add a verified email in Account security. Each verified email belongs to one RawBoard account. Pending registration details, including the salted password hash and agreement choice, are encrypted; no plaintext password is stored. Codes expire after ten minutes. Resending replaces the previous code. Pending requests are removed within 24 hours of the last request by the hourly maintenance check.

Resend receives your recipient address, verification code and email message to deliver an activation or email-change confirmation. These are transactional service messages, not marketing. Provider-held email records follow Resend’s retention and transfer arrangements described on the provider page. The server stores the verified address encrypted with a keyed lookup digest until you replace it or delete the account.

Optional two-step verification uses an authenticator app. Its setup key is encrypted on the server and shown to you during setup; the QR code is generated in your browser without an external QR service. Recovery codes are shown once and only keyed digests are stored. Each recovery code works once. Sign-in challenges expire after five minutes, and authenticator setup after ten minutes. Authenticators cannot reuse a previously accepted time-step code. Enabling or disabling two-step verification signs out other RawBoard sessions. Your trusted authenticator app has its own storage practices.

Short-lived, keyed security counters limit verification attempts and email sends. They contain no page content and expire with their ten-minute, one-minute or one-hour window. Account deletion removes the active email, authenticator secret, recovery records and account-bound challenges; managed backup expiry still applies. Account export includes your address, verification status and two-step status, and excludes secrets, recovery digests and challenge tokens.

Who is responsible

RawBoard is operated by Joels Rozkalns, Krogsils, "Mārstaļi", Ķekava, Latvia. Contact: joelrosehill@gmail.com. Telephone: +371 20667726.

This policy covers RawBoard’s released public website and account service at rawboard.site. Websites you visit have their own policies. RawBoard’s operator is responsible for the account and browser data described below.

Service usage and subscription statistics

RawBoard records named interactions with its own service to understand usage, subscription journeys, reliability and operating costs. These include account creation and successful sign-in, daily account activity, plan views and selections, checkout starts and errors, verified trial subscriptions and payments, renewal cancellation and withdrawal requests, confirmed withdrawal refunds, AI token and estimated cost totals, tab-open counts, Keep active use and service errors. A lightweight activity signal is sent while the app is visible; this is not a record of every click.

The administrator sees daily totals, charts and period comparisons. The analytics system does not collect the addresses or contents of visited websites, screenshots, typed answers, passwords, cookies, arbitrary interaction text, phone numbers, advertising identifiers or IP addresses. No analytics data is sent to an advertising or third-party analytics service, and no separate tracking cookie is added.

Daily activity and checkout records are linked to your RawBoard account for deduplication and purchase-journey comparisons for up to 90 days; they are included in account export and removed on account deletion. Aggregate daily counts and opaque deduplication keys are retained for up to 400 days. Existing financial and security retention rules apply separately. Managed backups expire under the backup policy. Historical statistics may include only records still retained at rollout, and client-side blocking can make optional interaction counts incomplete.

These first-party service statistics support the operator’s legitimate interest in understanding and improving the service and managing subscriptions. Turn optional usage statistics off in Privacy & account without losing features; linked activity and checkout-journey records are removed. Existing aggregate totals and necessary account, payment, AI allowance and security records remain. You can also contact the operator to exercise applicable data-protection rights, including an objection. Anonymous bug counters remain a separate system and do not contain account identifiers.

Trial verification

RawBoard does not request a mobile number or send verification SMS. Account activation uses email verification. Eligible new trials require subscribing to a plan through Stripe with a payment card; the trial lasts three days and the selected recurring price is charged afterward unless renewal is cancelled before the trial ends. Email, browser-marker and payment-card checks help prevent repeat trials.

The Firebase phone-verification flow has been retired. If a legacy verification receipt or security claim exists, it remains encrypted or pseudonymous under its original retention safeguards and is included in account export where applicable. Existing ten-minute verification challenges expire through maintenance; RawBoard does not send new phone-verification requests to Google.

Pages kept after closing RawBoard

Choosing Keep active until midnight transfers a compatible page to an account-isolated server browser. RawBoard processes its current page address, website-session cookies, available local/session storage, supported unsent form fields, viewport size and exercise-identification text to continue the page when the app is closed. A bounded list of navigation-control labels is held in device memory and transferred only when you keep the page, so a compatible popup exercise can be reopened. It excludes answer clicks, typed text and submission controls. Subsequent input in that view is forwarded to that server page. Ordinary device tabs keep executing on the device. RawBoard does not submit an exercise merely because you keep it.

Supported unsent fields and the reopening context are checkpointed in your encrypted account profile for recovery. Password, payment-card, hidden and verification-code fields are excluded from draft handover and checkpoints; this does not remove such information that a visited website independently stores in its cookies or browser storage. The server operator can access server-held data as described elsewhere in this policy. Keeping a page does not automatically send its content to an AI provider. Turn off the pin to return to the normal background expiry; account deletion removes the profile. Server or website failures can interrupt a kept page, and recovery depends on the same page and fields remaining available.

Free-trial abuse prevention

Eligible new users receive one free trial. To detect repeated claims, RawBoard stores keyed digests of the verified email address, a random signed browser marker and, for a subscription trial, the payment-card fingerprint returned by Stripe. RawBoard does not receive or store the full card number or security code. The fingerprint is transformed with a server-held key before being stored; it is not used for advertising. A repeated card trial is cancelled without silently converting it into an immediate paid purchase. Personal Gmail dot and plus aliases are treated as one mailbox; dots in school and custom email domains are preserved. These digests are pseudonymous security data, not a guarantee of anonymity. We do not obtain a MAC address, scan device hardware, use canvas fingerprinting or share these signals with an advertising provider.

The browser marker is an HttpOnly, SameSite cookie, Secure on HTTPS, with a lifetime of 180 days. Claim digests expire 180 days after first recording, including in the separately signed restoration ledger. Account deletion removes the account-bound eligibility decision; the limited claim digests can remain until that security period ends to prevent deletion and immediate trial re-registration. They contain no raw email address, IP address, username, page content or account ID. Other account records and provider retention follow their own schedules.

Short-lived keyed email, account, device and IP counters throttle bursts of registration, verification and API requests. Shared IP addresses are not a permanent account or trial ban. A device check can affect several people using the same browser: an account can still activate, buy a plan or redeem a valid friend code without a second trial. Contact joelrosehill@gmail.com for human review of a mistaken decision or to object to this security processing. The purpose is fraud prevention and service availability under the operator’s legitimate interest; these records are not used to decide school grades, profile learning ability or send marketing.

Automatic bug reports

RawBoard automatically reports detected application errors, failed page loads, suspected blank pages, unresponsive page commands and website-service interruptions to its own server to help repair the browser. Each report contains only a fixed technical error category, a RawBoard release identifier and a broad browser-engine category (Chromium, WebKit, Gecko or other). The server combines these into counts by UTC calendar day; it does not save individual reports, exact event times or a client identifier.

Bug-report payloads exclude names, account and tab IDs, IP addresses, page addresses and titles, page text or HTML, screenshots, form values, passwords, cookies, session tokens, AI prompts and answers, and raw error messages or stack traces. Reports are sent without account cookies or a referrer. No diagnostic identifier or queue is written to device storage, no advertising tracker is used and no third-party reporting service receives them. Duplicate categories are limited in device memory; failed delivery can receive one bounded retry. Ordinary successful browsing is not reported.

The reporting service necessarily receives network connection information, such as an IP address and HTTP headers, while delivering a request. It does not retain these in the reporting store, derive an identifier from them or link reports to accounts. This is a description of anonymous stored counters, not a promise that internet transmission itself is untraceable. Daily counters expire after 30 calendar days and are available only to authorized server administrators. Account deletion cannot identify or remove one person’s contribution to anonymous combined counts.

A suspected blank page can be an intentionally empty page, and a website script error may be harmless. Automatic detection does not identify every bug. A device/browser process that stops completely cannot send a report; reports may also be lost if connectivity does not recover or the app closes. Separate minimal server crash records contain component/error codes and filtered code locations, exclude page content and credentials, and follow the managed diagnostic retention schedule. Where transient connection processing is personal data, its purpose is service reliability under the operator’s legitimate interest; your rights remain available.

Information we handle

  • Your account: username, salted password hash, session records, creation date, versioned Terms acceptance, age-group and guardian-permission declarations, friend-code redemption and subscription status.
  • Your browser account: saved tab addresses and titles, website cookies, local storage, IndexedDB, saved website usernames and passwords, and recent page-assistant conversations.
  • AI and billing usage: token counts, estimated provider cost, allowances, plan periods, Stripe customer/subscription identifiers and payment status.
  • Technical information: connection and request information, errors and security or abuse-prevention events. Websites receive the information needed to handle browsing requests.

New account registration asks for a username, password and verified email address. A paid purchase or withdrawal asks for the purchaser’s full name and confirmation email. For an under-18 account, the purchaser must be an adult parent or legal guardian; their declaration and purchaser role are recorded with the contract. Contract copies, withdrawal notices and delivery status are retained separately from browser data. Stripe also collects payment and billing details during its own checkout. RawBoard does not receive your full card number or security code from hosted checkout.

Student tools and Office

Requested student tools read available page text, selected passages, grade tables and draft editors to prepare a result. Grade averages, simple draft formatting, timetable detection, calendar exports and document formatting run without sending that content to an AI provider. Grades and weights are editable; the result is an estimate, not an official semester grade.

Provider-assisted explanation, revision, research and tab organisation use reviewed, limited context and the existing privacy checks. Smart Clean sends only limited tab metadata; protected drafts, mail and grades are excluded from its shared snippets. It does not wake every sleeping tab. Word and Excel formatting stays in Office. An optional formula explanation shares the reviewed formula and request; cell values and workbook names are excluded and text literals are masked. The safeguards do not guarantee anonymity.

Saved tool results, study workspaces and tab groups are encrypted separately for your account and expire after 30 days. You can attach a workspace to a page: its reviewed notes, checklist and saved page titles then provide context for your AI requests. Saved page addresses are excluded from this shared workspace context, and attaching it does not open the saved tabs. Short-lived page context expires after 5 minutes, edit previews after 10 minutes and tab-closure undo after 24 hours. Expired records are removed by scheduled maintenance. Clearing AI history removes cached tool results; saved workspaces can be removed separately and account deletion removes these account records. Office connection tokens are temporary, scoped to formula assistance and last at most one hour. Calendar export creates a file on your device or opens an editable calendar form; it does not synchronise your calendar or send invitations. Draft tools never send email.

Why we use it

  • Providing the service: accounts, browser sessions, saved tabs and requested AI assistance are used to perform the service contract.
  • Subscriptions: usage and payment records support allowances, renewals and payment reconciliation, under the contract and applicable legal obligations.
  • Security: access checks, limited diagnostic records and spam controls protect accounts and service availability. Where GDPR applies, the basis is the operator’s legitimate interest in a secure service.
  • Requests and disputes: necessary information is used to respond to support, privacy and consumer requests and establish or defend legal claims.

Optional processing that legally requires consent must be explained separately. RawBoard currently has no marketing mailing list, advertising trackers or public-site analytics.

Page assistant and AI data sharing

The page assistant is for general questions, explanations, summaries and other requested help with the page you are viewing. It is not limited to exercises. When you ask AI, RawBoard sends the reviewed page context, your prompt and relevant recent conversation to DeepSeek so it can generate a response. Ordinary browsing does not itself send the page to DeepSeek. The API key remains on RawBoard’s server.

In the current implementation, context consists of extracted visible text, mathematical content, supported fields and answer choices. Text is limited and minimised before transmission. Cookies, saved website credentials, input values, page-link and image URLs, and URL query tokens are excluded. Recognisable email addresses, identifiers and credential-like text are redacted. Sign-in, personal-profile, inbox and payment-page checks block AI requests. No full-page HTML file or full-page screenshot is sent. An exercise diagram is sent as an original image only when you explicitly enable diagram sharing; text redaction does not redact an image. These are descriptions of current behavior, not guarantees that every private detail will be detected.

Server-side text filtering: Redact by Desert Ant Labs runs on RawBoard’s server in Latvia before each AI request, alongside deterministic credential and identifier checks. This is server-side processing; you do not need access to the operator’s private network or development environment. It replaces detected names, addresses and other identifying text with temporary aliases. Answers and supported insert actions are restored on the server. An unavailable, incomplete or timed-out filter stops the request; original text is not sent as a fallback. Unchanged text can reuse an account-isolated detection cache for up to five minutes. Deleting AI history, removing a website permission or deleting the account clears that cache. Cached text/detections and alias mappings stay in memory and are not written to filter logs. The text model runs within RawBoard’s infrastructure rather than a hosted text-redaction API.

Redact is a text detector, not a guarantee of anonymity or a detector of every confidential fact. It can miss identifiers or obscure details needed for an answer. Original images are not automatically redacted. Only the selected exercise diagram is eligible for optional image sharing, never a full-page screenshot. Desert Ant Labs receives the SDK’s required usage telemetry: a randomly generated runtime-installation identifier, application/SDK information, call counts and limited runtime context. RawBoard does not supply customer account IDs, prompts, page contents, detected identifiers or answers to that telemetry. Model inference stays on RawBoard’s server. The SDK’s license and telemetry terms apply. Per-request aliases are discarded after processing.

Website permission: you can choose to remember the AI sharing decision for an exact website origin and declare that you will use only permitted non-personal content there. It is saved for your own RawBoard account, separately from other accounts and other website origins. Later requests on that site can skip the context confirmation, including after page changes or signing in again. The same AI context-sharing prompt contains the optional lesson-image checkbox, alongside the content declaration and remember-for-this-website choice. You remain responsible for checking what you ask about; a saved choice does not certify the page as safe. Remove a permission in Privacy & account to restore confirmation. We store its encrypted origin, notice version, time and diagram preference until revoked or the account is deleted; your account export includes these choices. A new notice version requires a fresh choice.

When you enable image sharing, RawBoard can capture up to four relevant lesson images, SVG diagrams or canvas snapshots on your device. It excludes navigation and account headers; this is not a full-page screenshot. Choose lesson-image sharing in that same review prompt. To change a remembered choice, revoke the website permission in Privacy & account; the next uncached AI request shows the sharing prompt again. Images cannot be automatically anonymized. Captures are transient AI-request data, not anonymous bug reports or stored PDF files.

Page text, HTML and screenshots can contain personal information, even when the page is behind a login. Images are not anonymous merely because their content is shown as pixels. Review the context before sending it, and share only non-personal content you are entitled to share under the current provider arrangement. Do not include passwords, session tokens, payment details, sensitive personal information, children’s personal information or private information about another person without an applicable lawful basis. Skipping AI lets you continue ordinary browsing without that AI disclosure.

DeepSeek’s general privacy notice identifies processing and storage in China. Its API terms distinguish downstream-app end-user processing from the general notice. The public notice does not establish RawBoard’s API-specific retention or training terms. API-specific processing roles, training/reuse settings and the international-transfer arrangement remain unconfirmed. The API documentation describes default disk caching, with unused caches usually cleared within hours to days; this is not a guaranteed deletion deadline for all provider records. RawBoard makes no zero-retention, no-training or EU-only promise. See AI provider information and DeepSeek API caching documentation.

The registration checkbox records acceptance of the Terms and acknowledgment of this Privacy Policy. It is not blanket consent to share any private page, consent on behalf of other people, or a waiver of privacy rights. Requesting an AI answer instructs RawBoard to process the reviewed context for that answer; it does not replace any required lawful basis, processing agreement or international-transfer safeguard. Where processing relies on consent, the specific choice and an easy way to withdraw it must be provided separately.

RawBoard retains page-assistant conversation history for the period described below. Reopening a cached answer may avoid another API call; a follow-up question, explanation request or changed page may send context again and use more tokens. Any future full-page capture or additional AI feature that changes the data, purpose or recipient will be described before activation, with a new choice where legally required. Registration does not activate or approve those future data uses.

Storage, security and access

Device rendering: website JavaScript, page layout, canvas and manual input run in your device browser. RawBoard relays and rewrites network resources without running a website renderer. Each account and website has an isolated browsing origin, separate from RawBoard’s account area. The server retains encrypted website cookies, saved credentials, tab metadata and AI history; website local storage and IndexedDB are held on your device. Existing server profiles may retain older encrypted storage until removed. AI reads and explicit edits use an authenticated connection to the open device page. The provider key stays on the server.

The main server is located in Latvia. Browser profiles and saved website credentials are separated by account and encrypted at rest. Account passwords use salted hashes. Public connections use HTTPS. This is not end-to-end encryption: the server operator controls the system and encryption keys.

The administrator’s account screen shows usernames, subscriptions, token usage and estimated cost; it does not display other users’ browsing history or website passwords. Authorized operational access may still be necessary for maintenance, security or a verified support request. No system can guarantee that a security incident is impossible.

Detected games may run on your device or open separately, connecting directly to the game’s website. Third-party sites and payment/AI providers have their own roles and practices; see Service providers & data sharing.

How long information stays

AI chat history expires after 30 days. Managed database backups expire after 30 days through an hourly maintenance check. Existing undated AI chats receive a migration date once. Saved tabs, website storage and saved credentials remain while the account exists. A tab’s offload or focus expiry closes its renderer; it does not delete saved account data. Login sessions expire after 7 days; expired records are periodically purged.

Use Privacy & account to export your data, clear AI history or permanently delete your account after password confirmation. Deletion cancels renewal first and removes active account records, tabs, website sessions, saved passwords, AI history and the encrypted browser profile. Complete pending refunds before account deletion. Deleting RawBoard does not delete accounts on visited websites or Stripe’s independent records.

Minimal payment references, encrypted contract/withdrawal records and necessary confirmation-delivery records may remain for accounting and legal obligations. The default supporting-record period is at least five years, with unresolved disputes retained until resolved and longer legal obligations handled by the operator. These records exclude browsing history and website passwords. A minimal erasure ledger prevents managed backup restoration from reviving deleted accounts; historical managed backup copies expire within the managed backup period. Reviewed legacy configuration and database recovery copies have a separate 30-day expiry from their inventory date. Modified files and recorded holds require review; unknown copies and external exports are not automatically deleted. AI abuse-prevention request timestamps are pruned after their 24-hour budget window, while an active cooldown or recent strike can remain until it expires.

Your choices and rights

You can remove individual saved website passwords in RawBoard’s browser menu, manage website choices on the visited site, sign out, or cancel paid renewal in Subscription & tokens. To request access, correction, deletion, portability, restriction, or to object to processing based on legitimate interests, use Privacy & account for account export/deletion, or contact joelrosehill@gmail.com. The privacy request template tells you what to include.

Where GDPR applies, requests normally receive a response within one month; a permitted extension or refusal must be explained. Rights have legal exceptions. We may need proportionate information to verify account ownership, but you should never send a password or full card number. You can withdraw consent for processing based on consent without affecting earlier lawful processing.

You may complain to the data-protection authority where you live, work or believe an infringement occurred. European Commission: your data-protection rights.

Spam and allowance checks can automatically pause AI requests. They do not decide school results or credit eligibility. Contact the operator to ask for human review of an incorrect restriction.

Bugs, outages and service closure

Software bugs, maintenance, security incidents, network failures and changes to visited websites can interrupt RawBoard or affect saved state. The operator may temporarily restrict an affected feature while investigating an incident, protecting people or complying with a legal obligation. A takedown request is assessed according to its grounds and applicable law; receiving a request does not automatically determine the outcome.

If the service is closed, the operator will explain available account-data access or export options where reasonably possible and legally permitted, and apply the retention and deletion rules described above. Contact joelrosehill@gmail.com if an outage prevents you from exercising a privacy right. These events do not remove your rights to complain, seek a court remedy or claim compensation where the law provides it. Service availability and paid-plan remedies are explained in Terms of service and Cancellation & refunds.

Young users and changes

You must be at least 14 years old to have an account. If you are under 18, a parent or legal guardian must review and permit your use of the service. A paid plan for an under-18 account must be purchased by a parent or legal guardian aged 18 or above, using their own name and confirmation email. A school account or a friend code does not replace that permission. Do not send children’s personal information or sensitive personal information to the AI helper.

The date above identifies the latest policy update. Material changes to data use need an updated notice before the new processing begins and, where required, a separate consent request. Accepting service terms is not blanket consent for unrelated data use.