Last updated: October 14, 2026
Lecture Hat, Inc., a Delaware corporation ("Lecture Hat," "we," "us," or "our"), provides AI features as part of the Services covered by our Terms of Service. These additional AI conditions and processing disclosures apply when you use those features. "Input" means prompts, files, images, audio, and other materials submitted to an AI feature; "Output" means the material that feature generates. "Training" means using material to create or improve an AI model.
1.1 Our chat, summarization, knowledge-graph, transcription, captioning, and text-to-speech features are automated systems, not human beings.
1.2 You authorize us to process Input to provide the AI feature you use, including automatic indexing after upload and retrieval of relevant notebook content. You must have the rights and permissions needed to submit it. Authorized operators may access retained content for support, service operation, security, and legal compliance within the limits of your applicable agreement. Non-training review for product improvement must remain within permitted service-delivery and security purposes or have separate, specific permission; a description in the Privacy Policy does not itself grant that permission. Pilot research choices, Extension Limited Use restrictions, applicable law, and an executed DPA can further restrict access. We do not represent that every operator access is automatically audited.
The table describes the configured feature routes. Amazon Web Services, Inc. provides the AWS Bedrock routes. Claude requests use AWS global cross-Region inference and may be processed in supported commercial AWS Regions worldwide; Bedrock-routed Gemma requests use the Bedrock Mantle us-east-1 endpoint. Gemma in the standard hosted catalog runs Google Gemma 4 31B through AWS Bedrock. A deployment can separately enable a local inference option; when selected, that request is processed on Lecture Hat-controlled inference infrastructure. This does not make automatic indexing or other managed AI tasks local. Sections 3 and 9 state the training and retention limits; this is not a guarantee that inference stays in one country.
| Feature | Where it runs | What is sent |
|---|---|---|
| AI Chat and Q&A, including live-transcript chat and image understanding | AWS Bedrock for the standard hosted catalog, including Gemma and the Browser Extension; Lecture Hat-controlled inference for a separately enabled local chat option | Your prompt, text snippets retrieved from your notebook, any image you attach, and up to three retrieved page images. |
| Instant AI (highlight to explain) | AWS Bedrock (Claude and Gemma). Lead photographs are looked up at Wikimedia. | The highlighted phrase, the surrounding snippet of your own content, and your notebook's Instant AI instruction. Only the highlighted phrase goes to Wikimedia. |
| Inline writing suggestions (ghost text) | AWS Bedrock (Google Gemma 4 31B) | The text around your cursor, sent automatically after a pause in typing. You can turn this off in Settings, under Note Editor. |
| Practice-set generation, submission review, and teacher insight reports | AWS Bedrock (Claude and Gemma) | For generation and insight reports, your prompt and the content snippets the feature requires. For submission review, the assignment prompt and configuration, up to 80,000 characters of submitted work, selected submission metadata, up to 12 version-history snapshots capped at 1,200 characters each in the model prompt, and, only for classes configured to persist chat transcripts, up to 50 chat messages. Submission review runs after each accepted submission, including in Pilot and Strict Privacy classes; chat text is omitted for those classes. |
| Course Pilot chat (pilot-mode courses) | Same as AI Chat and Q&A | Your prompt and relevant content available to the course tutor, including course materials and notebook context. Additional terms apply to your course and are presented to you separately. |
| Document entity extraction during indexing | AWS Bedrock (Google Gemma 4 31B) | Extracted text and content snippets. Uploading a supported file starts this automatically. |
| Lecture summaries, document titles, handwriting transcription, and AI note composition | AWS Bedrock (Claude and Gemma) | The transcript, filename, handwritten page image, or note context the specific feature requires. |
| Live Subtitles (real-time lecture transcription) | ElevenLabs, Inc. (Scribe v2 Realtime) | An encrypted live audio stream for the session. Section 9 states the retention limits; Section 8.3 explains the unverified contractual coverage for institution-controlled data. |
| Text-to-Speech (Read Aloud and narrated lessons on web and iOS) | ElevenLabs, Inc. (Flash v2.5) | The response text or lesson script requested for playback or a narrated export, under the same provider limits stated for Live Subtitles. Preparing narration can fetch upcoming sections. The Browser Extension instead uses a voice installed on your device and sends no assistant text to Lecture Hat or to a third-party provider for Read Aloud. |
| Document OCR | Lecture Hat-operated PaddleOCR, with Amazon Textract as cloud fallback | Uploaded images and rendered PDF or Office document pages. If Lecture Hat-operated OCR is unavailable, page images are sent to Amazon Textract to extract text and its page coordinates. The entity-extraction row covers subsequent indexing. |
| Transcription of uploaded audio and video | Lecture Hat-operated Whisper, with ElevenLabs Scribe v2 as cloud fallback | Lecture Hat-operated Whisper is preferred; audio from the uploaded file is sent to ElevenLabs Scribe v2 when that service is unavailable. The ElevenLabs retention limitations and institutional restrictions below also apply. Subsequent indexing, summaries, and other AI features can separately send transcript text, extracted content, or relevant images to AWS Bedrock. Local transcription does not make the complete upload workflow local-only. |
| Synaptic Storage (knowledge-graph and search infrastructure) | Lecture Hat infrastructure for storage, retrieval, and embeddings | Your notes, transcripts, and uploaded files. Indexing can send extracted text and rendered page images to AWS Bedrock; chat and other AI features can also send retrieved content. Hosting, networking, and backup providers still apply to the stored records. |
Our Subprocessor List identifies providers and their roles, including storage, backup, and network delivery. Read it with this feature-level disclosure and our Privacy Policy. An executed institutional agreement controls its covered processing; a vendor list does not independently authorize a new use of your data.
3.1 Lecture Hat does not use your account content, including notes, files, audio, transcripts, chat messages, and knowledge graphs, to train, fine-tune, or otherwise improve any artificial-intelligence model, ours or a third party's.
3.2 AWS states that inputs and outputs submitted to Amazon Bedrock are not used to train the underlying models. AWS documents that content sharing with model providers is currently unsupported. Retention and AWS human review can depend on the effective mode, which we have not verified for every production request; this disclosure does not promise zero retention. Section 9 states the retention limits.
3.3 Our providers' terms may change. We will update this page if a change materially affects Section 3.2.
3.4 No permission for Training is granted by these AI conditions. Any future Training would require a separate affirmative opt-in describing the data, models, purposes, and withdrawal choices. Service debugging, research, and improvements that do not train a model are limited to the purposes and choices disclosed in the Privacy Policy and any executed DPA. Continued use is not consent to Training.
4.1 AI systems are not human and their Output may be inaccurate, incomplete, or out of date, including fabricated citations, misattributed quotations, and computational errors. Transcription can mishear speech or attribute it to the wrong speaker. Output may resemble material generated for others and may not qualify for intellectual-property protection. We do not guarantee its uniqueness or that it is free of third-party rights. The warranty provisions and mandatory consumer protections in the Terms apply.
4.2 Review Output before relying on or sharing it. Do not use AI for decisions in a regulated profession or for consequential academic decisions without qualified human oversight and compliance with applicable law and professional rules. Do not use it to infringe others' rights or falsely present AI-generated work as human-created. The Services are not a substitute for professional advice.
4.3 Many academic institutions restrict the use of AI tools on graded work. You are responsible for complying with your institution's academic-integrity policies. See our Acceptable Use Policy.
4.4 As between you and Lecture Hat, you retain your rights in Input. To the extent permitted by law, we assign to you any rights Lecture Hat has in your Output. That assignment does not transfer other people's rights or guarantee that copyright exists. These conditions do not reduce our duties under applicable privacy law or an executed DPA.
5.1 Our live-audio feature uses speech to produce a transcript; uploaded audio can also be stored and indexed. Lecture Hat does not create voiceprints to identify or authenticate you. We do not authorize that use by speech providers. Provider retention and model-improvement limitations remain as described in Section 9; this statement is not a guarantee about unverified provider account settings.
5.2 If we introduce a feature that uses voice as a biometric identifier, we will obtain your separate, informed, opt-in consent before activating it on your account and will publish the retention and destruction schedule required by applicable law.
5.3 You are responsible for obtaining any consent required by law before recording a lecture, class, or conversation. You acknowledge those requirements when you first use the recording feature.
6.1 Some AI features produce evaluative output: practice-answer scores, assignment-submission scores, teacher-facing review suggestions, and automated academic-integrity signals. All are advisory except the automated assignment score. Integrity signals may analyze text matching, similarity to up to 200 classmates' submissions, completion speed, differences between a coding submission and the student's prior code style, and use of AI chat. When automated assignment review succeeds, its 0 to 100 score is written to the submission's score field and the submission is marked reviewed. A model-identified verbatim paste sets that score within the 50 to 55 range. A teacher may override the score.
6.2 Lecture Hat does not determine academic standing and does not make admissions, employment, credit, insurance, housing, medical, or legal decisions. The product does not require human review before an AI-generated submission score is recorded. An institution or instructor decides how that score or other evaluative output affects a student and is responsible for qualified human review before any resulting academic decision.
6.3 Rights concerning automated decisions depend on applicable law and how the output is actually used, including current uses of assignment scores. Where required, we provide information about the processing and the means to seek an explanation, make representations, contest a decision, request human intervention, or exercise an applicable opt-out. A teacher's ability to override a score does not itself establish meaningful human review. Contact [email protected]; where an institution determines the decision, contact it as well and we will assist it with applicable requests. These rights do not depend on the introduction of a future feature. See the jurisdiction-specific rights and automated-processing explanation in the Privacy Policy.
Configured models as of October 14, 2026; availability can vary by deployment:
7.1 The model version routed to varies by feature and account tier and may change as providers release and retire models. The providers and the commitments in Section 3 do not change with the model version.
8.1 Chat, recording, and Read Aloud run only when you activate the corresponding control. Document indexing begins automatically on upload and may route extracted text to Google Gemma 4 31B through AWS Bedrock, page images to Amazon Textract for OCR fallback, or uploaded audio to ElevenLabs for transcription fallback. To avoid that processing, do not upload the document. Deleting an uploaded document does not undo processing that has already occurred. Permanent deletion enqueues asynchronous cleanup of derived index entries; it does not establish when all copies are removed. Inline writing suggestions can be turned off in Settings, under Note Editor.
8.2 We do not deactivate the ElevenLabs speech features automatically for any account type. A teacher may turn lecture transcription off for a class in the class settings. Provider-generated Read Aloud and lesson narration require Pro access.
8.3 We have not verified the applicable data-processing agreement and provider settings needed to approve ElevenLabs for institution-controlled data. Do not use Live Subtitles, provider-generated Read Aloud, narrated-lesson playback/export, or uploaded-media cloud transcription on content that your institution requires to remain under a data-processing agreement.
8.4 Contact [email protected] to discuss restricting provider processing. We do not offer a universal account-level switch for every provider. Until a restriction is confirmed and implemented, avoid the dependent features and uploads that automatically start processing.
9.1 When an enabled local inference option is selected, that inference request stays on Lecture Hat-controlled infrastructure. The standard hosted Gemma option uses Bedrock. Local inference does not imply that saved chat, operational logs, uploads, indexing, or other tasks are never retained or processed externally; the Privacy Policy and feature table explain those separate flows.
9.2 Claude requests use Bedrock Runtime and Gemma requests use Bedrock Mantle. AWS documents that retention and AWS human review depend on the model and effective account or project data-retention mode. Its current documentation says content sharing with model providers is unsupported. We have not verified the effective mode of every production Runtime or Mantle request. Read the AWS data-retention documentation; account configuration must be checked before relying on a zero-retention assurance.
9.3 Uploaded-media cloud transcription uses standard ElevenLabs processing with provider logging and retention enabled. We have not verified a fixed provider retention period. Our live-caption and speech-output requests ask the provider to disable logging, but we have not verified the effective Zero Retention Mode or model-improvement opt-out for our account. We do not promise zero retention or no provider model-improvement use; the applicable provider terms and account settings govern submitted audio and text.
9.4 Amazon Textract has separate terms from Bedrock. AWS permits storage and use of Textract content for service improvement unless an effective AI services opt-out applies. We have not verified that account-level setting or zero retention. See the AWS Service Terms and AI services opt-out documentation.
We may update the feature tables and processing disclosures as features, providers, or practices change. The "Last updated" date identifies the revision. We communicate material changes to processing and obtain any required consent as described in our Privacy Policy.
Changes to the additional AI conditions for standard Services follow Section 17.2 of our Terms of Service, including its advance-notice, cancellation, and refund rights. Changes affecting pilot participation remain subject to the applicable course terms and their acceptance requirements. An executed institutional DPA changes only through its agreed amendment process, and any required institutional approval must precede new processing. Updating this page alone does not amend an accepted course agreement or an executed DPA, or supply consent required by law.
The AI conditions supplement the Terms of Service; the feature tables explain current processing. Our Privacy Policy, applicable law, and any executed DPA govern personal information. An executed DPA controls covered processing in a conflict, and course-specific Pilot Terms govern activity within that pilot. Nothing here removes mandatory consumer rights or changes the liability and dispute provisions of your applicable agreement.
AI conditions adapted from the Common Paper AI Addendum, version 1.0, © Common Paper, Inc., under CC BY 4.0. Lecture Hat modified definitions, permitted processing, Training restrictions, ownership, and review obligations, and added its feature and provider disclosures. This is a customized adaptation; the unmodified standard terms are not incorporated by reference and Common Paper does not endorse it.