Built for classrooms. Answerable to IT.
This page says exactly what happens to student data, in plain language, with nothing aspirational. Share it with your IT or privacy office; they can hold us to every line.
Student data is never used to train AI models
We use enterprise API tiers of our AI providers (Google Vertex AI / Gemini). Student conversations and course materials are processed only to generate the immediate tutoring response. Neither we nor our providers train models on this data.
Answer keys teach the tutor without reaching students
Two controls, and they are not the same thing, so here is each one plainly. A student can never retrieve a material marked as an answer key: that is enforced in the retrieval code, not by a prompt anyone can talk around. The tutor itself is grounded in the key so it explains problems the professor's way, and every response is scanned server-side before it is sent, blocking any value that appears only in the key and nowhere the student can already look. That closes verbatim disclosure. It cannot stop a student inferring an answer from a correct explanation, and we will not claim otherwise.
Canvas access is read-only
Our Canvas LMS integration only reads course structure, assignments, and announcements that you choose to sync. We never write to your LMS, never modify grades, and never touch the gradebook. Disconnecting takes one click and revokes our access.
Encryption and access control
All data is encrypted in transit (HTTPS/TLS) and at rest on Google Cloud infrastructure. Student chat sessions are readable only by that student's account; professors see AI-generated class insights, not raw private conversations presented as surveillance.
Infrastructure you already trust
Encephlo runs entirely on Google Cloud (Firebase and Vertex AI), the same infrastructure many universities already approve. Payments for consumer users are handled by Stripe; we never see or store card numbers. Classroom students never pay at all.
Deletion on request, no data resale, no ads
We do not sell personal information, run advertising, or set first-party tracking cookies. Any student or professor can request full deletion of their account and data. We honor Global Privacy Control signals and applicable U.S. state privacy rights.
Documents for your IT office
Already written, already filled out, downloadable right now. No form, no email, no sales call in between.
Your own security questionnaire
If your institution uses its own review process instead, send it over. Answers come from the person who wrote the code, and they will match the documents above.
Straight answers for procurement
The questions IT offices actually ask, answered without spin.
Are you SOC 2 certified?
No, and neither is our HECVAT an official EDUCAUSE filing. We are an early-stage company, and we would rather say that plainly than put a badge wall between you and the facts. What exists today is above this line: a completed HECVAT Lite, a data-flow reference, a DPA, a subprocessor list, and a FERPA statement, all downloadable without asking us. Each one writes down what we do not have as carefully as what we do. The founder will also get on a call with your IT or privacy office. A formal audit is on the roadmap as institutional deployments grow.
Which third parties can see our data?
Six, and the subprocessor list above names each one and what it receives. Google Cloud runs the whole platform, including AI inference on Vertex AI. Instructure is your own Canvas, read-only. Microsoft and Apple appear only if a user picks that sign-in button. Resend delivers the weekly instructor digest, which carries class-level summaries and no student names. Inworld AI powers one of the read-aloud voices and receives only the text a signed-in user pressed play on, never a name or an account ID. Read-aloud also runs on Google's own speech service, so if your review would rather not add a vendor, that path involves no one you are not already trusting.
How does Encephlo relate to FERPA?
Encephlo pilots operate professor-first: professors invite students, students create their own accounts, and we act as a service provider processing only the data needed to tutor. We do not receive institutional education records unless your institution chooses to provide them under an agreement. For institutional deployments we will sign your data processing agreement and work within your FERPA framework.
Which AI models process student questions?
Google Gemini models via Vertex AI enterprise endpoints. Prompts include the course materials the professor uploaded and the student's conversation; responses stream back to the student. No student data is retained by the model provider for training under the enterprise terms we use.
What do professors see about their students?
Professors see class-level insights: what topics students ask about, where the class is struggling, engagement over time, and per-student summaries oriented around who needs help. The design goal is teaching signal, not surveillance.
Who do I contact for a security review?
Email suraj@encephlo.com. You will get the founder directly, not a ticket queue. We are glad to complete questionnaires, walk through the architecture, and sign reasonable agreements for pilots.
See also our Privacy Policy and Terms of Service.