Beta notice — not yet legally reviewed. This page describes how Second Read Studio actually works during a small public beta. It has not been reviewed by a lawyer and is incomplete until the items below are supplied by the operator.
Still required from the operator:
- Confirmation of applicable legal bases
- Confirmation of hosting regions and processor agreements (Lovable, Supabase, Render, OpenRouter, Google, Ahrefs)
- Verified retention periods for hosting logs, backups, AI providers and privacy-request records
Privacy
Last updated 4 November 2026
Who is responsible
The operator and data controller is Kapsalis Christoforos. For privacy questions or requests, contact cikapsal@secondread.studio. The outstanding processing details above still need to be confirmed.
Summary
Second Read Studio uses Google sign-in and Supabase to keep private account history. When you save a draft or start a review, its title, text, editorial context and rubric selection are stored in your account. Validated reviews, suggested edits, accept/reject decisions and current explanations are stored so you can continue later without another AI request. Old browser-local work is not automatically imported.
Optional public website analytics
With your analytics permission, Ahrefs also counts visits to the home, support, privacy, and terms pages. Its script runs in an isolated frame with only the public page address and a fixed site title. It cannot access the Studio page content. Query strings, fragments, referrers, account identifiers, form submissions and link clicks are not passed to Ahrefs. Ahrefs receives browser language, screen dimensions and connection metadata, including your IP address. The frame is removed when you withdraw consent or enter a private page. This installation measures public page visits only and does not provide referral attribution. See Ahrefs privacy policy.
Beta applications
Google Forms collects your sign-in email and intended writing use so we can manually review your application and contact you about beta access. A public LinkedIn or X profile URL is required and is reviewed only for that purpose. Applying does not subscribe you to marketing. Unsuccessful applications are deleted within 90 days of the decision, including any response-sheet copies. You can request earlier deletion by email.
Google sign-in and private cloud history
Supabase manages your sign-in session. Your account profile may contain Google’s display name and avatar, while Google and Supabase process sign-in information under their own policies. Access rules restrict drafts and history to their owner. A trusted server verifies ownership before storing a validated AI review. Signing out clears the app’s cached account data, but does not delete your stored writing.
You can permanently delete an individual saved document from History. This also removes its saved reviews, decisions, versions, edit evidence, project wiki sources and notes from that document, and document-specific rubric selection. Deleting a document does not delete your account or profile.
Saved account data remains until you delete it or request erasure; there is currently no automatic inactivity expiry. Supabase’s policies are available at Supabase privacy.
Your data and your choices
Account & privacy provides a readable Markdown writing playbook and a full JSON archive, plus permanent account deletion after a recent sign-in. The playbook combines the editorial foundations used by Second Read with your saved writing context, approved personal rules, rubric sets and a descriptive history of your recorded choices, including recent feedback notes you left on decisions; it does not include your draft text. The JSON archive contains the underlying drafts and account records. Choice patterns describe recorded activity, not a measured change in writing skill. These controls do not require beta approval. Deletion removes the login and associated profile, documents, versions, reviews, decisions, rubric selections, projects, project wiki sources and notes, preference proposals and learning records from the active database. It also invalidates refresh sessions. It does not delete your Google account.
Account deletion is irreversible. Requests already sent to model providers cannot be recalled. Provider logs, hosting logs and backup copies are separate from the active database and may remain until their retention periods expire. We have not yet verified all those periods. Copies you downloaded or stored on other devices remain under your control. Clear this site’s browser data on those devices; open tabs may continue displaying cached text.
You can request access, correction, erasure, restriction of processing, portability, or object to processing by emailing the privacy contact above. This also covers data outside the self-service export, such as beta applications, correspondence and provider records. We aim to respond without undue delay and within one month. Where legally permitted for a complex request, an extension of up to two further months requires notice and reasons within the first month. Proportionate identity verification may be needed; do not email passwords or sensitive drafts. You may complain to your local data protection authority, including the Hellenic Data Protection Authority.
Purposes and legal bases
The proposed basis for account storage and reviews you request is providing the service under Article 6(1)(b) GDPR. Optional preference learning uses your consent under Article 6(1)(a), which you may withdraw in Editorial Profile. Necessary abuse prevention and security use legitimate interests under Article 6(1)(f), subject to a documented balancing assessment. The operator must confirm these bases and the assessment before treating this notice as final. Editorial suggestions do not make legally or similarly significant decisions about you.
Recipients and international processing
Lovable hosts the application; Supabase provides authentication and account storage; Render runs the review and learning service; OpenRouter routes model requests to Google. Google also handles sign-in and beta application forms. Fonts are served by this site. These services may process connection metadata such as IP addresses. Provider roles, hosting regions, processor agreements and safeguards for transfers outside the EEA remain to be verified. An app feature cannot replace those agreements or verification.
Your writing setup is stored privately in Supabase. The saved answers may inform optional preference learning, but do not become instructions automatically. “Use as default for projects” controls whether these answers prefill new projects. Project settings apply to their new drafts, while existing documents keep their saved context.
Optional personal preference proposals
Learning is off by default. If you enable it, a separately configured private weekly worker may send bounded editorial decision summaries, feedback explanations and short manual before/after fragments to OpenRouter and Google Gemini. Full manuscripts are not sent by this worker. At least 10 actions across two documents are required. You may also explicitly request one smaller proposal from saved questionnaire answers using the same opt-in and weekly limit.
Proposals, supporting evidence IDs, your approval/rejection/disable decisions and learning run metadata are stored privately in Supabase. Only rules you approve can guide subsequent matching reviews; document-specific context and Second Read’s factual integrity take precedence. Turning learning off stops future analysis but does not disable previously approved rules or recall provider submissions already in flight. You can disable rules and delete pending proposals in Editorial Profile. Account deletion removes the remaining stored learning records; use the privacy contact for provider-side follow-up.
Learning uses the same OpenRouter data-collection denial, not a verified Zero Data Retention guarantee. Model costs and failure codes for learning are stored persistently in Supabase; this is separate from Render’s temporary review counters. Learning and scheduling may remain unavailable until the operator completes deployment.
What happens when you start a live review
The author’s short Project context can be included in a review. If project memory is enabled, up to four active core notes for that project are included as source-linked context. A separate model selection call receives your draft and an index of up to 24 other notes, and may choose up to three to open for that review. The official project page and full earlier documents are not fetched during reviews. Reviews saved before this change may have included short same-project phrases or a public page extract; their historical metadata is preserved.
Optional project wiki
If you enable the wiki for a document, saving changed draft text or changing its project setup starts a separate compilation. Its saved draft and up to 20 active notes from your own documents with the exact same Project context are sent to Render, OpenRouter and Google Gemini. The compiler may also read the single public HTTPS page you supplied. That website receives a normal page request, never your draft. Compiling uses a model call; merely opening History, comparing versions, restoring, or resuming a document does not compile it.
The source text, any fetched page extract, compiled notes, short verbatim evidence quotes, and your accept/reject choices are stored privately in Supabase. New notes become active automatically; you can reject them. The wiki may also compile a project-specific writing convention that a saved draft or uploaded file states explicitly. Those conventions guide later reviews for the same project but remain subordinate to the Second Read skill, the current draft and your document constraints. Notes from a draft are the writer’s claims, not independently verified facts. A document’s wiki sources and notes are deleted when that document is deleted. Project files and notes follow their project, and account deletion removes the full wiki and project setups.
You can also add a Markdown, text, PDF, Word or PowerPoint file to an enabled project wiki. The file is sent to our Render service to extract selectable text. That extracted text is sent to OpenRouter and the model for a separate wiki compilation. We store the extracted text, filename and source-linked notes in Supabase, but do not retain the original binary file. Removing a project file deletes its extracted text and compiled notes. Image-only pages are not read unless you convert them to selectable text first.
Your title, draft text, editorial context and enabled custom rubrics are sent from this site’s server to our review service (hosted on Render), and from there to OpenRouter, which forwards the request to the model Google Gemini 3.8 Flash, served by Google. Model configuration can change during the beta; the provider metadata is included in saved review data. The review that comes back is checked and returned to your browser. Our service does not write your text to disk or logs.
OpenRouter and Google process your text under their own terms and privacy policies, which govern how long they retain it: OpenRouter privacy, OpenRouter terms, Google privacy.
Provider data-collection setting — not Zero Data Retention
Each request asks OpenRouter to route only to providers that do not collect or train on submitted data (OpenRouter’s data_collection: deny setting), with no fallback to other models.
This is not Zero Data Retention. We do not currently request OpenRouter’s Zero Data Retention routing, and we have not verified that your text is never retained. Providers may keep it for limited periods, for example for abuse monitoring, as their policies describe. Do not submit confidential, personal or sensitive material.
What our service records
The Render inference service logs and counts only technical facts: a random request ID, character and paragraph counts, number of rubrics, outcome (success or error code), duration, token counts and model cost. It never records draft text, titles or rubric text. These counters live in memory and reset whenever the service restarts. This is separate from the intentional private document and review storage in Supabase.
Browser storage
- Account reviews and decisions: saved in Supabase, with temporary page memory while you work. They are not written to local or session storage by the new account workflow.
- Sign-in session: Supabase stores credentials in browser storage to restore your session. Signing out removes the session and clears cached account data.
- Legacy anonymous live reviews and decisions: old tab-session records remain separate and disappear when that tab closes. They are not imported into an account.
- Demo reviews’ decisions and saved rubric sets: this browser’s local storage. They stay until you clear them (History → Clear, or your browser settings). Rubric sets you save stay on this device.
- History tags: stored in this browser’s local storage under your account ID. They do not sync to another browser or device and are not sent to Supabase.
- Anonymous visitor cookie (
sr-visitor, HttpOnly, 30 days): a random ID used only to apply fair-use limits on live reviews. It is not linked to your identity and contains no draft content.
New draft text remains on the page until you choose Save & close or Start review. Both save it to your account. Reader-question answers are optional and remain in page memory only.
Usage limits
Live reviews are limited per anonymous visitor and per day, and total model spending is capped with the provider. Limits are best-effort and may change during the beta.