Privacy

Prereasoner analyzes the spreadsheet data and questions you submit so it can compute an answer and show the steps used to reach it. The reference service at chat.prereasoner.com is operated by MailRecipe LLC.

What we process

• Files and spreadsheets you choose. In the Google Sheets™ add-on, this is limited to the visible, non-empty tabs in the current spreadsheet, including table names, columns, and cell values. They are read and sent to a conversation kept for that spreadsheet when you open the add-on, so a question can start at once, and again if they have changed when you ask. The Excel add-in does the same with the visible, used ranges in the open workbook, including values and column labels; filtered or hidden rows in a visible worksheet are included. It does not edit cells. That conversation expires like any other, and is not listed among your conversations until you ask a question in it.
• Your questions, conversation history, generated SQL, results, and reasoning traces.
• Account and session identifiers supplied through Google, Microsoft, and Firebase Authentication.
• For the Google Sheets™ add-on, the document ID needed to reopen the conversation associated with that spreadsheet for your signed-in account.
• For the Excel add-in, a one-way hash of the workbook URL is used to reopen its associated conversation; the workbook URL itself is not stored for this purpose.
• Operational data needed to secure, debug, and operate the service.

How answers are produced

The core engine builds and executes inspectable SQL against your data. It uses a frozen model as an encoder for intent and schema signals; no model on the service's servers writes SQL or invents the numerical answer, and the database computes it. When the engine cannot build a query on its own, it can ask Gemini for help as described below, and the answer says so.

When the engine cannot build a query for a question on its own, it can ask Gemini to reword the question once. That request contains the question, the table and column names, column types and foreign keys. It contains no example cell values or conversation history. The deterministic typed search builds and validates every query, and the database computes every number. Values you include in your question are part of that question.

Starter questions are generated from sheet names, column names, the active sheet and selected source scope only. No cell values, formulas, computed answers or conversation history are sent for suggestions. Only visible, non-empty sheets are described, and a suggested question names only sheets and columns that exist; selecting it fills the question box for your review.

Reference-table generation is a separate, explicit action. When you request it, Gemini receives the reference table's headers, entity names, existing reference cells and your instruction so it can fill missing reference cells. These generated reference values require your review.

Google receives those requests through the service's Google Cloud project. Its handling is governed by the operator's Google Cloud agreement and is also described in Vertex AI data governance and the Google Cloud Privacy Notice. Prereasoner is working toward operating this assistant layer locally once a local model meets the required quality, latency, security, and cost standards. The calculation and verification path remains deterministic.

Service providers

• Google Cloud and Firebase: hosting, authentication, databases, and reasoning-trace delivery. See Firebase privacy and security.
• Google Cloud Vertex AI (Gemini): language processing for the hosted assistant features and the optional wording rewrite described above.

Microsoft sign-in for Excel uses Microsoft's standard identity response through Firebase Authentication. Prereasoner receives the resulting Firebase account/session identity; it does not call Microsoft Graph or store a Microsoft access token.

Why we process data

We process this data to provide the analysis you request, maintain your account and saved work, secure and debug the service, and meet legal obligations. Depending on the law that applies, the bases are performance of the service contract, legitimate interests in operating and protecting the service, and compliance with legal obligations. We do not treat a protocol field or a popup click as blanket permission for unrelated uses.

Google API Limited Use and AI/ML

We use Google user data only to provide the user-facing spreadsheet analysis, saved conversations, and related security and support functions described on this page. We do not use, transfer, or sell raw, aggregated, anonymized, or derived Google user data to create, train, or improve generalized or non-personalized artificial-intelligence or machine-learning models. Data sent to Gemini on Vertex AI is used only to process the user's requested feature; it is not sent or authorized for Google or third-party model training.

Sharing and international processing

We do not sell customer spreadsheet data or use it for targeted advertising. We share data with the providers above only to operate the service, and may disclose information when required by law or needed to protect the service and its users. Those providers can process data in the United States and other locations where they operate. We use the applicable contractual safeguards when data-protection law requires them.

Data protection

Prereasoner uses administrative, technical, and organizational safeguards designed to protect Google user data against unauthorized access, use, alteration, disclosure, loss, or destruction:

• Data is encrypted in transit using HTTPS/TLS. Application data stored in Google Cloud SQL and Firebase is encrypted at rest by Google Cloud's managed encryption controls.
• Every application request requires a short-lived, server-verified Firebase identity token. Server-side ownership checks isolate each user's conversations, and Firebase rules limit reasoning-trace reads to the authenticated owner.
• Cloud SQL is reached through an IAM-authenticated connector rather than an open database network. Runtime services use dedicated, least-privilege service accounts and a non-superuser database role.
• Database credentials are stored in Google Secret Manager and are not embedded in client code or operational logs. Gemini is called with the service's own Google Cloud identity, so no model API key exists.
• Request-size, row, rate, and storage limits reduce abuse and unnecessary data exposure. Production logs exclude spreadsheet rows, questions, conversation history, generated SQL, credentials, request bodies, and full error messages.
• Human access to user content is prohibited except with the user's explicit permission for support, when necessary to investigate abuse or a security incident, or when required by law.

Storage and retention

Uploaded tables are stored in a conversation-scoped PostgreSQL schema. Conversation state and named workbook revisions, and spreadsheet conversation links are stored in application tables with verified user ownership. Conversations and Sheets conversation links expire after 90 days without activity by default. The service also limits how many conversations and how much stored source and workbook state each account keeps. So that an answer whose delivery failed can be sent again, each finished answer is kept for ten minutes and then removed. Reasoning traces stored in Firebase Realtime Database expire after seven days by default. A daily retention job applies these policies.

Operational logs contain operation names, sizes, and error classes. They do not contain raw request bodies, questions, conversation history, generated SQL, credentials, spreadsheet rows, or full error messages.

Deletion

Deleting a conversation removes its owned PostgreSQL data and the reasoning traces indexed to that conversation, and every answer the service is keeping to resend. Clearing all conversations deletes each conversation owned by the signed-in user in the same way, one at a time; a reasoning trace not indexed to any conversation expires with the other traces after seven days. Deleting only the Firebase identity does not currently erase every PostgreSQL or trace record. For complete account-linked deletion, email mani.doraisamy@gmail.com.

Training and public data

Customer spreadsheet rows and data derived from them are not included in Prereasoner's training corpora and are not used to train or improve shared models. The frozen encoder used by the service is not updated from customer data. Public reference facts come from documented sources such as Wikidata and publisher datasets, under their respective terms.

Your choices and rights

You can delete conversations in the product and request access, correction, export, restriction, objection, or deletion where applicable by emailing the address below. Do not upload regulated health, payment-card, or similarly sensitive data unless your deployment has a specific agreement and controls for it.

Children

The reference service is not directed to children under 18. Contact us if you believe a child has submitted personal data so we can investigate and remove it where appropriate.

Contact

MailRecipe LLC
340 S Lemon Ave #9974, Walnut, CA 91789, United States
Privacy and data-rights requests: mani.doraisamy@gmail.com.
Security reports: see the security policy.