Subprocessors
Version 2026-10-25 · Last updated 25 October 2026
These are the third parties that can receive data from TutorTalks, what each one gets, and why. A service that is not on this list does not receive student content from us. Websites where your institution embeds a session, and any tool servers it connected itself, belong to you, not to us; they are covered below.
Providers that can receive student content
Student content means what a student says and writes in a session, their voice, and the records made from it. The defaults today: OpenAI handles every AI step of a live conversation, Google Gemini drafts and edits sessions for instructors, HeyGen streams every video avatar, and ElevenLabs is used only if we select it.
The region is where each provider mainly processes data for us. Providers may also process data in other countries under their own terms and safeguards.
| Provider | Purpose | What it receives | Region | Model training |
|---|---|---|---|---|
| OpenAI (default) | By default: speech recognition, the tutor's replies, the tutor's voice, knowledge search (embeddings), the coach and conversation notes that steer the tutor, and Student insights reports. Also speech to speech, where an instructor turns it on for a session: the tutor model hears the student's speech itself and answers aloud. Also the AI Session Designer and AI edit, if we switch them to OpenAI. | Student speech audio. The conversation text and the tutor's instructions, which include the name the student typed at join and their pronoun answer. The student's words, to search the session's knowledge base. Excerpts of instructor material. Reply text to be spoken. The transcript, to write an insights report. Never the student's email address. | United States | Not used to train OpenAI's models by default. OpenAI may keep API data for up to 30 days to monitor for abuse. |
| Google (Gemini API) | By default, the AI Session Designer and AI edit, which draft and revise a session from an instructor's brief and uploaded material. We can select it instead of OpenAI for the tutor's replies, the coach and notes, insights reports or knowledge search. | For the Designer and AI edit: the instructor's brief, uploaded files and the session being edited. No student records; the session's list of approved student emails is never shown to it. If selected for a conversation role: what OpenAI would receive for that role, including the student's name and pronoun answer. | Google's infrastructure (no region is set) | Paid API service: Google does not use what we send to improve its products. Google keeps it for a limited period to detect misuse. |
| HeyGen (LiveAvatar) | Streams the video avatar. It is the only avatar provider and is used by every video session; voice-only sessions do not use it. | The tutor's reply audio, to animate the avatar, and which avatar to show. The participant's browser connects directly to HeyGen's real-time video service, so HeyGen sees that browser's IP address and details. The student's microphone is never sent to HeyGen, and no name or email field is. Instructors' browsers also load avatar preview images from HeyGen. | United States | We have opted out of HeyGen's use of our inputs to improve its models. |
| ElevenLabs (optional) | Speech recognition or the tutor's voice, only where we select it instead of OpenAI. Not the default. | Student speech audio, for speech recognition; reply text to be spoken. No name or email field, although a reply can address the student by name. | United States | We have opted out of ElevenLabs' use of our content to improve its models. Its zero-retention mode is not used. |
| Google Cloud / Firebase | Database (Firestore), transcript store (Realtime Database), file storage (Cloud Storage), instructor sign-in (Firebase Authentication, which also sends password-reset emails) and receiving payment events from Stripe (Cloud Functions) | Everything TutorTalks stores: accounts, sessions, student session records, transcripts, audio when recorded, insights reports and access logs; instructors' sign-in details. Students never sign in with Google. | United States (Firestore multi-region nam5; Cloud Functions us-central1) | No. Google stores and processes it for us as our cloud provider. |
| Hostinger | Hosts the TutorTalks application server, which every session passes through, and our internal operations console | Live audio and conversation text, in memory, while a session runs. Uploaded avatar media. Server logs, kept for 30 days, which record IP addresses and the pages requested, including any search text in a page address (such as a student's name or email address an instructor searches for), but never transcripts or audio. Temporary export files, deleted once sent. | Boston, Massachusetts, United States (application server) | No. Infrastructure provider only. |
Providers that do not receive student content
| Provider | Purpose | What it receives | Region |
|---|---|---|---|
| Stripe | Payments, subscriptions, usage billing and invoices | The buyer's name and email address; a billing address (required at checkout); an organization name and tax ID, if given; card, bank or wallet details, entered directly into Stripe's own pages (they never reach us); our account reference, the buyer's TutorTalks sign-in ID and, for purchases made through Stripe Checkout, the reference of the terms-acceptance record and the versions of the Terms and Privacy Policy accepted; plan, invoice and metered-usage amounts. No student data of any kind. | United States and other countries where Stripe operates |
| Google Analytics (via Firebase) | Measuring how the pricing, credit-redemption, instructor and studio pages are used. Never loaded on the student join, session or embed pages. Those same pages load Google's Firebase sign-in library from Google's servers. | Pages viewed and basic interactions, browser and device details, approximate location derived from the IP address, and an identifier kept in cookies or browser storage. We do not use it for advertising. No student data. | Google's infrastructure |
| jsDelivr | A content delivery network that serves two open-source code libraries to the browser: three.js on student, instructor and studio pages, and the LiveKit client in video-avatar sessions | The IP address and browser details of anyone whose browser loads those pages, students included. No conversation content. | Worldwide (nearest server) |
| Transactional email provider | Where email sending is configured: one-time credit links when we grant credits, and welcome emails with a sign-in link for accounts we set up. Account notices we email, such as notice of a price or terms change, may also go through it. | The recipient's email address and the message. No student data: we do not email students, and team invitations are handed to the person who sent them rather than emailed by us. | Depends on the provider |
Websites and servers your institution connects
Websites where you embed a session
On the Plus plan and above, an instructor can embed a session in a page on their own website. That website is yours, not our subprocessor. It cannot read inside the session's frame. The frame tells the host page only that it has loaded (and how tall it is), that the student has joined, with the session's title, that the session has ended, and error codes. It never passes the student's name, email address, transcript or audio.
Anything the host page runs itself, such as its own analytics, is the responsibility of whoever operates that website. An instructor can limit which websites may embed a session; if none are listed, any website can.
Tool servers (Model Context Protocol): withdrawn
TutorTalks contains a feature, withdrawn from sale and hidden in the dashboard since September 2026, that lets a session call a server you run or subscribe to (a booking system, a lab database, your own API) while a conversation is happening. It is not offered to new customers, but an account that set up tool servers before the withdrawal may still have them in use. Those servers are not our subprocessors. Nobody at TutorTalks chooses them or has an agreement with them; under FERPA they are your institution's to authorise, as with any other system you connect a course tool to.
Such a server receives only the arguments the tutor composes for a tool call. Those are written from the conversation, so they can contain what a student said. It does not receive the transcript, the audio, the student's email address or the session's instructor material. Servers are reached over HTTPS only, tools are allowed by name, session by session, and every session starts with none allowed.
Providers and engines not in use
Earlier versions of TutorTalks could animate an avatar with NVIDIA Audio2Face-3D, or run speech recognition (Whisper) and avatar animation on our own servers. None of these is used today, so NVIDIA is no longer listed. If that changes, this page will change first.
What we require of every provider
- Business terms, not consumer apps. We use each provider's business or API service, and we have opted out of the provider's use of our data for model training wherever the provider offers that choice. The tables above say what each provider's terms allow.
- Only what the job needs. A student's email address is never sent to an AI provider. The language-model provider does receive the name a student typed at join and their pronoun answer, because the tutor uses them. The name field is optional; a student who leaves it blank is recorded as Anonymous.
- Retention as each provider states it. A provider keeps what it receives only as its own terms allow; OpenAI, for example, may keep API data for up to 30 days to monitor for abuse. We do not promise zero retention at any provider.
- No advertising. We have no advertising partners, and we do not sell personal information or share it for advertising.
- Security commitments. We use providers that publish security and incident-notification commitments, and we tell affected institutions if a provider's incident involves their student records.
Notice of changes
Before a provider that is not on this page begins receiving student content, we update this page and tell the owners of institution accounts by email. Moving a job between providers already listed here is an operational setting; the first table says which provider is the default for each job. If your institution's agreement with us gives you a right to notice or to object, that agreement governs.
Contact
TutorTalks is operated by Arclight Innovation LLC. Questions about this document, about a student record, or about a security concern go to studio@arclightxr.com.
If you are a student: requests to see, correct or delete your education records go to your school, not to us. Your school decides those questions; we carry them out on its instruction. Your instructor or your registrar's office is the right first contact.