Privacy Policy

ThinkTorch AI LLC (“ThinkTorch,” “we,” “us”) operates the ThinkTorch website and teacher tools at https://thinktorchai.com.

ThinkTorch is designed for educational writing feedback. It is not designed for healthcare, counseling, medical-record storage or management, mandated reporting, or student-information-system functions.

This Privacy Policy describes how we handle information in the current invite-only U.S. teacher beta. Access is limited to teachers and education professionals we provision. There is no public self-service signup. There are no student accounts.

The live production-ready teacher product in this beta is Writing Feedback Assistant. Other screens on the site may be previews, planning experiments, or unfinished tools. This policy focuses on the hosted service as it actually operates today, especially Writing Feedback Assistant.

The service is in active use with invited teachers, but it is still a beta. Features, storage practices, and this policy may change. We will update this page when they do.

This policy is meant to be readable. It is not a claim of certification, district approval, FERPA or COPPA compliance, or a signed data-processing agreement.

ThinkTorch AI LLC does not sell teacher personal information or student data.

Depending on how you use the service, you may provide:

• Account information. An email address we use to invite you, send a sign-in code, and operate your teacher session.
• Waitlist or contact information. If you join the public waitlist or email us, the address and message you send.
• Assignment context. Assignment directions, rubric text, teacher notes, grade/course labels, genre, and Writing Feedback Assistant settings you enter.
• Student writing. Ordinary classroom writing that you paste or type into Writing Feedback Assistant so the tool can generate academic feedback and run Student Safety Review.
• Optional student name or identifier. A label you may enter for your own on-screen report, copy, or Word filename. You do not have to provide one.
• Safety follow-up actions. If Writing Feedback Assistant asks you to acknowledge or record a Safety outcome, the fact of that action and any notes the product is designed to keep as operational records.

You decide what you submit. We process what you send in order to provide the service.

Writing Feedback Assistant is built for teachers to submit ordinary classroom writing. Submitting student work is the purpose of the tool.

Before submitting real student work, confirm that your school or district permits use of ThinkTorch or external AI services for that purpose. If you are unsure, check with the appropriate administrator, technology representative, or privacy representative. Using ThinkTorch, or accepting these terms, does not override school or district policy.

Avoid including unnecessary direct identifiers in drafts or assignment context, such as full student names, student ID numbers, email addresses, phone numbers, or home addresses, unless you genuinely need them and are authorized to include them.

The optional Student Name or Identifier field is for teacher-facing report labeling. In the current product:

• that value is not sent directly to the Academic model or the Safety model; and
• if you enter a value in that field, ThinkTorch removes exact matches of that value from the text sent to ThinkTorch server and model paths before processing.

The original draft remains visible to you in your browser. Report labels may still use the name or identifier you entered.

This is not a claim that ThinkTorch removes all personal information, finds every name or identifier in student writing, or fully de-identifies submissions. Other names, places, contact details, or identifying details that you did not enter in that field are not automatically removed.

When you generate feedback, ThinkTorch typically does the following:

1. Identifier step (if used). If you entered a Student Name or Identifier, exact matches of that value are removed from model- and server-bound text as described above.
2. Academic review uses an AI model to generate teacher-facing academic feedback. The student draft, plus assignment directions, rubric text, and teacher notes if you provided them, are sent for that academic review.
3. Student Safety Review separately uses an AI model to review the submitted draft for potentially concerning language. The draft is sent separately for that review. Safety does not receive your rubric, trait selections, or the optional student name/identifier field.
4. Zero Data Retention routing. Both model requests are configured through Vercel AI Gateway to require Zero Data Retention (ZDR) routing. That means ThinkTorch asks the Gateway to use only providers that qualify for ZDR for that request. If no ZDR-compliant provider is available, the model request is designed to fail rather than continue on a non-ZDR route.
5. Return to you. Academic feedback and Safety results are shown in your browser session so you can review them. One Generate produces one report, including Safety as part of the same review. Suggested skill levels are not automatically grades. “Possible Inappropriate AI Use” is not proof of authorship or misconduct.

ThinkTorch does not name a single permanent upstream model host in this policy, because Gateway may use any provider that satisfies the ZDR requirement.

ZDR routing is a configuration ThinkTorch applies to these model requests. It is not a promise that no technical logs or metadata can exist anywhere, and it is not a guarantee of every technical behavior of third-party infrastructure beyond the configuration we actually control.

Student Safety Review draws your attention to potentially concerning language in the submitted writing. It is not a diagnosis, a threat assessment, a legal determination, or a mandated-reporting determination. ThinkTorch does not make mandated reports on behalf of teachers or schools. You review the actual student writing, apply professional judgment, and follow applicable school or district procedures.

ThinkTorch does not automatically send Safety flags to parents, schools, law enforcement, child-protection agencies, or other third parties solely because a Safety flag occurs. ThinkTorch may disclose information if required to comply with applicable law or valid legal process.

ThinkTorch does not intentionally persist in its application database:

• the raw student draft after processing
• the generated academic feedback text
• Safety-detected passages or surrounding excerpt text

Word reports are generated on our server when you ask for them and are not intentionally stored there. Plain-text copy uses your device clipboard.

Application logs are not written to include raw student drafts.

“Not intentionally persisted” is not a claim that a draft can never appear in a transient processing buffer, an error path, or a provider system during the request.

We do retain limited operational records needed to run, secure, and account for the service. That currently includes:

• teacher account and authentication information
• Writing Feedback Assistant usage and report-accounting information (for example, that a Generate used a report)
• Safety operational metadata, such as a submission hash, alert level, review identifiers, timestamps, and related metadata — not the draft itself and not Safety passages
• records of product use needed to debug failures, prevent abuse, and operate the beta

We do not publish a fixed deletion clock for these records, because the product does not currently enforce one.

• Authentication cookies / session. Signed-in teachers use hosted authentication. Session cookies keep you signed in while you use the teacher tools.
• Saved Assignment Context. If you save assignment context in Writing Feedback Assistant, your browser may store directions, rubric text, teacher notes, and settings locally. The student draft is not deliberately saved in that Writing Feedback Assistant local storage. The optional student name/identifier is not saved there either.
• On-screen labeling. The optional student name/identifier may remain in the page for report labeling during that session and may be sent transiently if you generate a Word file, so the document can be labeled.

Clearing site data, using another browser, or signing out can affect what remains on your device.

On the hosted service, we use Vercel Analytics for product and page analytics (for example, which pages are used). It is not a student-work archive, and it is not used to store drafts.

We use other companies to operate ThinkTorch. They are not ThinkTorch employees. They process information as needed to provide their services. Categories include:

• Hosting and application infrastructure, including Vercel
• Authentication and database, including Supabase
• AI model / inference infrastructure, including Vercel AI Gateway, which routes Academic and Safety requests under the ZDR configuration described above
• Analytics, including Vercel Analytics
• Email delivery used to send sign-in codes or operational messages

Our service providers operate under their own terms and contractual obligations. ThinkTorch configures Academic and Safety model requests to require Zero Data Retention routing as described above. ThinkTorch does not guarantee every technical behavior of third-party infrastructure beyond the configuration we actually control.

We use reasonable administrative, technical, and organizational measures appropriate to a small hosted beta, including authentication for teacher tools and access controls on operational data.

No online service is perfectly secure. Do not treat ThinkTorch as a student-information system of record, a counseling record, or a substitute for your school’s required reporting channels.

Raw Writing Feedback Assistant drafts are processed to generate results and are not intentionally kept by ThinkTorch after that processing.

Operational, security, usage/accounting, and Safety-metadata records may be kept as reasonably needed to run the service, prevent abuse, and meet legal obligations.

There is no self-service Delete Account control in the current product.

To ask about your account or data, email christian@thinktorchai.com. We will review requests in good faith. We do not promise immediate deletion, complete erasure of every backup or log, or deletion of records we must keep.

ThinkTorch provides tools to invited teachers. Your school or district — not ThinkTorch — decides whether you are authorized to submit student work, what student information may be used, and how results may be shared with students or families.

If you are not authorized to submit a particular draft, do not submit it.

This beta is for authorized teachers and education professionals. It is not offered as a student product and does not provide student accounts.

We may update this Privacy Policy. Updated policies will be posted on this page.

For material changes that affect how teacher or student information is handled, ThinkTorch will provide reasonable notice to active beta users where practicable.

Continued use of the service after an update takes effect is subject to the revised policy. If you do not agree, stop using the service and contact us.

ThinkTorch AI LLC
https://thinktorchai.com
christian@thinktorchai.com