Back to Trust Centre
Auditable AI Gateway & Model Disclosure
AI Transparency
Last updated: 26 September 2026 · Every AI feature the platform can run, the model behind it, and who receives the content.
Executive AI Transparency Summary · EU AI Act Alignment
Personal Academy operates an auditable AI Gateway; training on customer data is excluded only where the provider offers it, as recorded per provider on the sub-processor page. Every capability (syllabus synthesis, auto-grading rubrics, Yvion copilot) is bound to our database feature registry, metered per action, and protected by hard tenant monthly quotas. Enterprise customers can configure Bring-Your-Own-Key (BYOK) for dedicated billing and isolation.
How this page is produced
Generated from the platform’s AI feature register, not written by hand. A feature that is not listed here cannot run: the gateway refuses any AI activity that has no register entry, and a database trigger refuses to configure one. Configuration shown is the platform default — an organisation may point a feature at a different model, in which case its own administrators see that in AI Configuration.
AI features
59 features. “Sent to the provider” is the whole of what leaves the platform for that feature.
| Feature | Who uses it | What it does | Sent to the provider | Model |
|---|---|---|---|---|
| Connector event matchingconnector_event_matchaddon.ambient | Platform operatorsInternal self-test; no customer content | Decides which learning goal, if any, a connected-tool event relates to, so that a learning moment can be offered. | A minimised event descriptor (type, title keywords) and the list of learning goal titles. No message bodies. | deepseek/deepseek-v4-flashvia openrouter |
| Learning momentlearning_momentaddon.ambient | LearnersThe model drafts content a person then reviews | Writes a short, timely learning prompt for a learner based on an event in their flow of work.Users are told: “This learning prompt was written by AI based on your recent work activity. It is a suggestion, not an instruction.” | A minimised description of the work event and the learner role and learning goals. No message bodies or private content. | deepseek/deepseek-v4-flashvia openrouter |
| Meeting briefmeeting_briefaddon.ambient | LearnersThe model drafts content a person then reviews | Prepares a short preparation brief for a learner ahead of a meeting, drawn from their learning goals and the meeting title.Users are told: “This meeting brief was written by AI from your calendar entry and learning goals. Review it before you use it.” | The meeting title and time, attendee count, and the learner learning goals. No meeting notes or attendee personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Admin Assistantadmin_assistantmodule.ai | AdministratorsThe user converses with the model directly | Conversational assistant for platform administrators. It answers questions about the AI agents configured on the platform and, at the administrator's instruction, changes those agents' settings (prompts, model, temperature, top_p, max tokens, enabled switch).Users are told: “You are chatting with an AI assistant. It can view this platform's AI settings and apply the changes you ask for; every change it makes is recorded in the version history and can be undone.” | The administrator's typed request, the current page they are on, and a summary of the AI agent settings (names, models, temperatures, token limits). No learner personal data is sent. | openai/gpt-4o-minivia openrouter |
| AI learning coachai_coachmodule.ai | LearnersThe user converses with the model directly | Answers a learner's questions about the lesson they are reading, explains concepts a different way and sets practice questions.Users are told: “You are talking to an AI. Replies are generated by a language model and can be wrong — check anything that matters against the lesson itself.” | The lesson title and lesson text, and the messages the learner types. | openai/gpt-4o-minivia openrouter |
| AI mentor check-inmentor_checkinmodule.ai | LearnersThe user converses with the model directly | Asks a learner how their learning is going and suggests next steps.Users are told: “You are talking to an AI. This is generated advice, not a message from a person.” | The learner's own stated goals and their course progress summary, and the messages they type. | openai/gpt-4o-minivia openrouter |
| Ask Yvionyvion_sales_agentmodule.ai | prospectThe user converses with the model directly | Conversational assistant and coach across public and authenticated pages: explains the platform, assists learners and educators, and guides visitors.Users are told: “You are chatting with Yvion, an AI sales assistant - not a human. It can explain the platform and arrange a follow-up; it cannot make commitments on the company's behalf. Your conversation is stored so it can follow the thread and is deleted automatically after a limited period (90 days by default). If you share contact details, they are passed to our team so someone can follow up.” | The visitor's typed messages, the recent turns of their own conversation, and the marketing page they are reading. No account, learner, tenant or other person's data is sent. | deepseek/deepseek-v4-flashvia openrouter |
| Ask Yvionplatform_agentmodule.ai | Platform operatorsThe user converses with the model directly | Conversational assistant and coach available throughout the platform. It explains concepts, navigates to learning materials, and proposes actions for the user to confirm.Users are told: “You are chatting with an AI assistant. It can look things up on your behalf and suggest actions; nothing changes until you confirm it, and every confirmed action is recorded in the audit log as made via the assistant.” | The person's typed message, the page they are on, their permissions and entitlements, counts of items needing their attention, retrieved help text, and the results of read tools they are allowed to run. No other person's personal data is sent unless the caller is allowed to read it. | deepseek/deepseek-v4-flashvia openrouter |
| Homepage AI sandbox (public site)hero_ai_sandboxmodule.ai | prospectThe model drafts content a person then reviews | A one-shot demonstration on the public homepage: a visitor types a short training request and sees the kind of draft the platform would produce (a course outline and a knowledge check). It never makes decisions about a person, stores nothing about the visitor, and never sees tenant or account data.Users are told: “Demo - simulated. This reply is an AI demonstration, not a human, and it is generated by a simulated provider. It is illustrative only and nothing you type here is kept.” | The short request the visitor typed (at most 400 characters). No account, learner, tenant or other personal data is sent. | hero-sandbox-demovia mock |
| L&D consultantconsultant_advisemodule.ai | AdministratorsThe user converses with the model directly | Advises an L&D lead on their programme, reading the organisation summary the platform already shows them.Users are told: “You are talking to an AI. This is generated advice, not a recommendation from a named adviser.” | A summary of the organisation's courses, enrolments and completion rates, and the messages the adviser types. | openai/gpt-4ovia openrouter |
| Learn anything (from your sources)learn_groundedmodule.ai | LearnersThe model drafts content a person then reviews | Writes a short workplace explainer for a learner using only passages retrieved from the organisation knowledge base and courses, and cites the passages it used.Users are told: “This explainer was written by AI from your organisation's documents. Check the cited sources before relying on it.” | The topic the learner typed and the retrieved passages the learner is allowed to read. No other personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Learn anything (general knowledge)learn_generalmodule.ai | LearnersThe model drafts content a person then reviews | Writes a short workplace explainer for a learner from general knowledge when nothing in the organisation content covers the topic and the learner chose this. It never describes an organisation policy and carries no citations.Users are told: “This explainer was written by AI from general knowledge, not from your organisation's documents. Check before relying on it.” | The topic the learner typed. No organisation documents and no other personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Learning Playground exerciseplayground_generatemodule.ai | LearnersThe model drafts content a person then reviews | Builds one practice exercise (roleplay, case, quiz drill or code task) on the fly from a skill the learner chose, their job title, department and open competency gaps. Scored by the existing roleplay/quiz scorers; the result becomes skills evidence.Users are told: “This exercise was generated by AI for your practice. Scores here are indicative.” | The chosen skill, exercise type and difficulty; the learner's job title, department and open gaps (names only, no scores of other people). | deepseek/deepseek-v4-flash-0731via openrouter |
| Pipeline diagnosticecho_testmodule.ai | Platform operatorsInternal self-test; no customer content | Built-in self-test. Returns a fixed echo of the assembled prompt without calling any provider — used to prove the pipeline works. | A test message typed by a platform administrator. No learner content. | mock-echo-v1via mock |
| Reading companionreading_companionmodule.ai | LearnersThe user converses with the model directly | Explains a passage of lesson content on request, at the learner's pace.Users are told: “You are talking to an AI. Replies are generated by a language model and can be wrong — check anything that matters against the lesson itself.” | The passage of lesson content the learner selected, and the messages they type. | openai/gpt-4o-minivia openrouter |
| Roleplay feedbackroleplay_scoremodule.ai | LearnersThe model scores or comments on something a person produced | Reads back a completed roleplay transcript and gives the learner feedback against the objectives the author set.Users are told: “This feedback was generated by an AI from your practice conversation. It is practice feedback only — it does not decide anything about your record.” | The scenario, objectives and the full transcript of the practice conversation. | openai/gpt-4o-minivia openrouter |
| Roleplay partnerroleplay_responsemodule.ai | LearnersThe user converses with the model directly | Plays the other person in a practice conversation — a customer, a colleague, a patient — so a learner can rehearse.Users are told: “The character you are speaking to is an AI, not a person. The conversation is generated by a language model.” | The scenario, the persona and objectives the author wrote, and the learner's turns in the conversation. | openai/gpt-4o-minivia openrouter |
| Socratic tutorsocratic_tutormodule.ai | LearnersThe user converses with the model directly | Guides a learner toward an answer with questions and hints instead of giving the answer directly.Users are told: “You are chatting with an AI tutor, not a person. It may make mistakes, so check anything important with your instructor.” | The learner messages in the current tutoring conversation and the lesson topic. No other personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Talking roleplay (voice)roleplay_voice_turnmodule.ai | LearnersThe user converses with the model directly | Turns the learner's spoken reply in a practice conversation into text and speaks the AI persona's answer in a synthetic voice.Users are told: “The voice you hear is AI-generated and the character is an AI, not a person. Your recording is transcribed and then discarded; only the text is kept.” | The learner's recorded reply (sent to the speech provider, not stored) and the persona's reply text. | scribe_v1via elevenlabs |
| Process feedbackprocess_feedbackmodule.assessment | LearnersThe model drafts content a person then reviews | Gives a learner written feedback on the reasoning steps of their submitted work against the assessment rubric.Users are told: “This feedback was written by AI against the assessment rubric. Your instructor makes any grading decision.” | The learner submission, their recorded working steps and the assessment rubric. No other personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Question bank generatorassessment_question_generatemodule.assessment | Course authorsThe model drafts content a person then reviews | Drafts assessment questions for an author to review and add to a question bank.Users are told: “Drafted by AI. Review every question before adding it to the bank.” | The source material the author selected and the question types requested. | deepseek/deepseek-v4-flash-0731via openrouter |
| Rubric grading assistantassessment_rubric_suggestmodule.assessment | AdministratorsThe model scores or comments on something a person produced | Suggests a rubric score and comments on a submitted answer, for a human grader to accept, change or ignore.Users are told: “This suggested score was produced by an AI. A grader's decision is what counts.” | The rubric the author wrote and the learner's submitted answer. | Registered, not configured on this deployment — it cannot run |
| Viva question generatorviva_questionsmodule.assessment | Course authorsThe model drafts content a person then reviews | Drafts oral-examination style follow-up questions for an author to review, based on a learner submission or an assessment topic.Users are told: “These questions were drafted by AI. Review and edit them before you use them with learners.” | The assessment topic or submission text and the rubric. Learner identity is not sent. | deepseek/deepseek-v4-flashvia openrouter |
| AI narration (text-to-speech)studio_narration_generatemodule.authoring | Course authorsThe model drafts content a person then reviews | Synthesizes spoken-word audio from a lesson script or an Audio block's narration text, for use in course lessons.Users are told: “This audio was AI-generated (text-to-speech) from the script shown alongside it. Review it before publishing -- AI-generated narration must be disclosed to learners, the same as AI-generated imagery.” | The spoken script text (lesson content or an author-written narration script) and the chosen voice/model. No learner personal data. | elevenlabs/eleven-v4via openrouter |
| Block intent assistantblock_intentmodule.authoring | Course authorsThe model drafts content a person then reviews | Turns one lesson block into what the author asks for: a worked example, a scenario check, a job aid, a simpler version or industry examples.Users are told: “Suggested by AI. Review before saving.” | The one block, its lesson title and nearby text, the course learning pattern and the chosen intent. | claude-sonnet-5-5via anthropic |
| Content block refinerrefine_blockmodule.authoring | Course authorsThe model drafts content a person then reviews | Rewrites one content block — shorter, longer, clearer — on the author's instruction.Users are told: “Suggested by AI. Review before saving.” | The text of the block and the author's instruction. | openai/gpt-4o-minivia openrouter |
| Course module plannergenerate_modulesmodule.authoring | Course authorsThe model drafts content a person then reviews | Proposes the module structure for a whole course.Users are told: “Suggested by AI. Review before saving.” | The course title and description. | openai/gpt-4ovia openrouter |
| Cover image briefcover_briefmodule.authoring | Course authorsThe model drafts content a person then reviews | Turns a title, description and audience into three short visual briefs for a cover or thumbnail image. The image itself is produced by the studio_image_generate activity. | The item's title, description, audience and kind (course, path, programme, lesson). | deepseek/deepseek-v4-flash-0731via openrouter |
| Field enhancerenhance_fieldmodule.authoring | Course authorsThe model drafts content a person then reviews | Rewrites a single course field — a title, a description — on the author's instruction.Users are told: “Suggested by AI. Review before saving.” | The field being edited and the course title for context. | openai/gpt-4o-minivia openrouter |
| ID consultant: assessment designerid_consult_assessmentmodule.authoring | Course authorsThe model drafts content a person then reviews | Drafts quiz items from the course objectives at the matching Bloom level, with distractor rationale and a difficulty spread, saved as draft questions in the question bank.Users are told: “These questions were AI-generated drafts. Review, edit and approve them in the question bank before use.” | The course objectives, their evidence criteria and the lesson titles. No learner personal data. | claude-sonnet-5-5via anthropic |
| ID consultant: blueprint critiqueid_consult_blueprintmodule.authoring | Course authorsThe model suggests tags or labels a person accepts or rejects | Critiques a course blueprint against its brief and design pattern (coverage, cognitive load, sequence, practice density, assessment alignment, time) and can propose an alternative design under a different pattern.Users are told: “This critique was AI-generated from your blueprint. You decide what to change.” | The course brief, the blueprint and the chosen design pattern. No learner personal data. | claude-sonnet-5-5via anthropic |
| ID consultant: brief interrogationid_consult_briefmodule.authoring | Course authorsThe model suggests tags or labels a person accepts or rejects | Interrogates a course brief as an instructional-design consultant: five questions, a rewritten performance need, an audience card, a verdict on whether training will solve the problem and a recommended design pattern.Users are told: “This advice was AI-generated from your brief. You decide what to change.” | The course brief the author wrote and the list of learning design patterns. No learner personal data. | claude-sonnet-5-5via anthropic |
| ID consultant: learner persona testid_consult_personamodule.authoring | Course authorsThe model suggests tags or labels a person accepts or rejects | Reads a lesson as a named learner persona (novice, expert, sceptic, non-native reader) and reports confusion points, reading level, time estimate and a verdict.Users are told: “This is a simulated learner reaction generated by AI, not feedback from real learners.” | The lesson blocks and objectives and the persona name. No learner personal data. | claude-sonnet-5-5via anthropic |
| ID consultant: lesson critiqueid_consult_lessonmodule.authoring | Course authorsThe model suggests tags or labels a person accepts or rejects | Critiques one lesson for learning quality (one idea per block, example before practice, feedback specificity, reading level, inclusive language, alt text, placeholders) and proposes replacement blocks.Users are told: “This critique was AI-generated from your lesson. Review every suggested block before applying it.” | The lesson blocks, the lesson objectives and the chosen design pattern. No learner personal data. | claude-sonnet-5-5via anthropic |
| Lesson content generatorgenerate_contentmodule.authoring | Course authorsThe model drafts content a person then reviews | Writes the body of a lesson from an author's brief.Users are told: “Drafted by AI. Review and edit before publishing.” | The author's brief and the surrounding lesson context. | openai/gpt-4ovia openrouter |
| Lesson enhancerlesson_enhancemodule.authoring | Course authorsThe model drafts content a person then reviews | Rewrites the wording of a lesson's existing content blocks on the author's instruction, keeping every block, its type and its order. The result is a proposal the author reviews and applies.Users are told: “Proposed by AI. Review each change before applying.” | The text of the lesson's content blocks and the author's instruction. | openai/gpt-4o-minivia openrouter |
| Lesson transcriptionlesson_transcribemodule.authoring | Course authorsThe model drafts content a person then reviews | Transcribes the speech in an audio or video file an author uploaded to a lesson and builds a caption track from the timed transcript.Users are told: “This transcript and its captions were generated automatically from the lesson audio. They may contain mistakes; the author can correct them.” | The audio track of the uploaded lesson media file. No learner personal data is sent to the provider. | scribe_v1via elevenlabs |
| Module lesson plannermodule_lessonsmodule.authoring | Course authorsThe model drafts content a person then reviews | Proposes an ordered list of lessons for a course module.Users are told: “Suggested by AI. Review before saving.” | The course title, module title and intended duration. | openai/gpt-4o-minivia openrouter |
| Studio artifact generatorstudio_artifact_generatemodule.authoring | Course authorsThe model drafts content a person then reviews | Drafts facilitator guides, workbooks, agendas and session plans from the approved blueprint.Users are told: “This training material was AI-drafted from your approved blueprint. A facilitator must review it before delivery.” | The approved blueprint summary (objectives, modules, activities, timings). No learner personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Studio assessment generatorstudio_assessment_generatemodule.authoring | Course authorsThe model drafts content a person then reviews | Proposes draft assessment questions from a blueprint objective's evidence criteria for the author to review in the question bank.Users are told: “These questions were AI-drafted from the objective's evidence criteria. They are saved as DRAFTS in the question bank — nothing is approved or scored until a human reviews and approves each item.” | The objective statement, its evidence criteria and lesson context. No learner personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Studio blueprint generatorstudio_blueprint_generatemodule.authoring | Course authorsThe model drafts content a person then reviews | Drafts a course outline, objectives and activities from the author's brief and selected source excerpts for the author to edit and approve.Users are told: “AI helped draft this blueprint from your brief and selected sources. Every objective, module and activity is a suggestion you review and approve before anything is generated from it.” | The course brief, selected source excerpts, delivery modes, duration budget and language. No learner personal data. | openai/gpt-4ovia openrouter |
| Studio image generatorstudio_image_generatemodule.authoring | Course authorsThe model drafts content a person then reviews | Generates an illustrative image from the author's visual brief for use in course lessons.Users are told: “This image was AI-generated from your visual brief. Review it, add alt text, and confirm it suits the content before use. AI-generated imagery must be disclosed to learners.” | A text visual brief written by the author (subject, style, mood) and the course locale. No learner personal data. | google/gemini-3.1-flash-lite-imagevia openrouter |
| Studio lesson generatorstudio_lesson_generatemodule.authoring | Course authorsThe model drafts content a person then reviews | Drafts lesson content as editable blocks for an approved blueprint lesson.Users are told: “This lesson draft was AI-generated from your approved blueprint and sources. Review, edit and approve the content before publishing.” | The lesson title, objectives, duration, mode and the selected source excerpts for that lesson. No learner personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Studio lesson translationstudio_translatemodule.authoring | Course authorsThe model drafts content a person then reviews | Translates a Studio lesson's block content into a target locale, preserving block ids and types so learner state carries over.Users are told: “This translation was AI-generated. A reviewer who reads the target language must approve it before it is published to learners.” | The lesson title, source and target locale codes and the lesson's block content. No learner personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Studio video clip generatorstudio_video_generatemodule.authoring | Course authorsThe model drafts content a person then reviews | Generates a short illustrative video clip from the author's visual brief for use in course lessons.Users are told: “This video clip was AI-generated from the author's visual brief. Review it in full and confirm it suits the content before use. AI-generated video must be disclosed to learners.” | A text visual brief written by the author (subject, motion, style) plus optional duration/resolution parameters. No learner personal data is sent to the provider. | minimax/hailuo-3-maxvia openrouter |
| Community post polishpolish_postmodule.community | Course authorsThe model drafts content a person then reviews | Tidies a community post's wording at the author's request.Users are told: “This post was refined by AI at the author's request.” | The post title and body the author wrote. | openai/gpt-4o-minivia openrouter |
| Mentor match suggestionmentor_matchmodule.community | AdministratorsThe model suggests tags or labels a person accepts or rejects | Suggests which of up to five anonymous candidate mentors best fits a mentoring request, with a one-sentence reason. A person who manages programmes approves, changes or declines every suggestion; nothing starts without that approval.Users are told: “AI suggested this mentor from anonymous summaries. A person decides whether the match goes ahead.” | The mentoring goal the mentee typed (PII-scrubbed), the names of the skills they want to grow, and for each candidate only a job title, a department, the shared skill names and spare capacity. No names, ids or email addresses. | deepseek/deepseek-v4-flashvia openrouter |
| Tag suggestersuggest_tagsmodule.community | Course authorsThe model suggests tags or labels a person accepts or rejects | Suggests tags for a community post. | The post title and body. | openai/gpt-4o-minivia openrouter |
| Drip message writerdrip_generatemodule.engage | Course authorsThe model drafts content a person then reviews | Writes the scheduled nudge messages for a drip campaign.Users are told: “Drafted by AI. Review before scheduling.” | The course title, module name and the day number in the campaign. | Registered, not configured on this deployment — it cannot run |
| Ask your data (question routing)data_question_routemodule.insights | AdministratorsThe model suggests tags or labels a person accepts or rejects | Matches an administrator question about organisation learning data to one fixed report template and its parameters. The figures come from the nightly rollups, never from AI.Users are told: “AI chose which report answers your question. The figures come from your dashboards, not from AI.” | The question the administrator typed and the current date. No learner data is sent. | deepseek/deepseek-v4-flashvia openrouter |
| Report analysisreport_analysismodule.insights | AdministratorsThe model drafts content a person then reviews | Reads the aggregated rows of a report the caller can already see and returns a short narrative, up to three anomalies with the rows behind them, and suggested next actions that link to the pages that act.Users are told: “AI-generated reading of this report. Check the figures before acting on them.” | The report name, its column labels, up to 200 aggregated rows (never more than the caller may read) and an optional question. | deepseek/deepseek-v4-flash-0731via openrouter |
| Cited answer from the knowledge basecited_answermodule.knowledge | LearnersThe user converses with the model directly | Answers a learner question using only passages retrieved from the organisation knowledge base, and cites the sources it used.Users are told: “This answer was written by AI from your organisation documents. Check the cited sources before relying on it.” | The learner question and the retrieved knowledge passages. No other personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Knowledge base indexingknowledge_index_embedmodule.knowledge | Platform operatorsInternal self-test; no customer content | Creates search embeddings for knowledge base documents so that answers can be retrieved and cited. | Chunks of text from documents the organisation has added to its knowledge base. No learner personal data. | openai/text-embedding-3-smallvia openrouter |
| Flashcard generatorflashcard_generatemodule.learn | LearnersThe model drafts content a person then reviews | Turns a lesson into revision flashcards.Users are told: “These flashcards were generated by an AI from the lesson content.” | The lesson title and lesson content. | openai/gpt-4o-minivia openrouter |
| Lesson content translationcontent_translatemodule.learn | Course authorsThe model drafts content a person then reviews | Translates a published lesson's block content into another language, preserving block ids and types so learner quiz progress carries over.Users are told: “This translation was AI-generated. A reviewer approves it before learners see it.” | The lesson title and block content (answer keys are never included). | deepseek/deepseek-v4-flashvia openrouter |
| Course recommendation re-rankingrecommend_rerankmodule.paths | LearnersThe model suggests tags or labels a person accepts or rejects | Optionally re-orders an already-computed, rule-based course recommendation list; never selects courses on its own. | The learner's top candidate courses (titles plus the rule-based reasons already computed). No personal data. | deepseek/deepseek-v4-flashvia openrouter |
| Personal learning plan generationplan_generatemodule.paths | LearnersThe model suggests tags or labels a person accepts or rejects | Selects and orders courses from a pre-computed candidate pool (skills gaps plus recommendations) into a personal learning plan; cannot introduce a course outside that pool. | Competency gap names and levels (no raw scores), candidate course titles, and the learner's stated weekly time budget. | deepseek/deepseek-v4-flashvia openrouter |
| Product help (AI guide)platform_helpmodule.platform | Platform operatorsThe user converses with the model directly | Answers how-to questions about the product from the help centre, opens the right page, and hands the person to a human when it cannot answer from the help text. Free to the customer; the platform pays for it within fair-use limits.Users are told: “You are chatting with an AI guide, not a person. It answers from the help centre and can hand you to a person; nothing changes until you confirm it.” | The person's typed question, the page they are on, their permissions and entitlements, retrieved help text, and the results of read tools they are allowed to run. No other person's personal data is sent unless the caller is allowed to read it. | deepseek/deepseek-v4-flashvia openrouter |
| Skill suggesterskills_suggestmodule.skills | Course authorsThe model suggests tags or labels a person accepts or rejects | Suggests which skills a course teaches, for the author to accept or reject. | The course title and description. | openai/gpt-4o-minivia openrouter |
AI sub-processors
Who receives content when an AI feature runs, and what each has committed about training on customer data. We state a commitment only where we can cite it.
| Provider | Sub-processor | In use by default | Training on customer data | Notes |
|---|---|---|---|---|
| Anthropic APIanthropic | Anthropic PBC | Yes | Not verified — we have not confirmed this provider’s current position, so we do not claim one | Not verified. |
| Azure OpenAI Serviceazure_openai | Microsoft Corporation | Available, not used by default | Not verified — we have not confirmed this provider’s current position, so we do not claim one | Not verified. |
| Built-in mock providermock | None — no network call | Yes | Not applicable — no content leaves the platform | No network call is made. The response is generated locally from the assembled prompt and nothing leaves the platform. |
| Customer-configured webhookwebhook | Endpoint configured by the customer | Available, not used by default | Not verified — we have not confirmed this provider’s current position, so we do not claim one | The receiving endpoint is chosen by the customer, so the platform cannot state who processes the content or on what terms. |
| Google Generative AIgoogle | Google LLC | Available, not used by default | Not verified — we have not confirmed this provider’s current position, so we do not claim one | Not verified. |
| OpenAI APIopenai | OpenAI, L.L.C. | Available, not used by default | Not verified — we have not confirmed this provider’s current position, so we do not claim one | Not verified. |
| OpenRouteropenrouter | OpenRouter, Inc. | Yes | Not verified — we have not confirmed this provider’s current position, so we do not claim one | Not verified. OpenRouter routes to an upstream model provider, so a second sub-processor is implied by the model chosen. |
Controls a customer holds
Opt out. An organisation can switch off all AI features, or any single feature, from its own settings. The switch is read on every AI request rather than stored in a session token, so it takes effect on the next request rather than when a token expires. A switched-off feature refuses and says which switch is off; it does not fall back to a lesser answer.
Human confirmation. A feature marked “consequential” above cannot produce a final answer about a person. Its output is recorded as a recommendation and stays that way until a named person with theai.decision.review permission confirms, overrides or rejects it — with who, when, and any note kept. No feature is marked consequential today.
Your own model. An organisation can supply its own provider key, per module, in which case calls for that module go to its account rather than the platform’s.
Every text AI call passes one gateway that checks permission, this register, the organisation’s opt-out and quota, applies mandatory safety instructions, and records the call — feature, model, provider, tokens and cost — in the organisation’s own usage log and audit trail. Image, video and narration generation pass the same permission, register, opt-out and quota checks, reserve their credits before the provider is called, and are recorded in the same usage log.
Generated live from the AI feature register. Questions a security review needs answered and this page does not cover? Email support@personalacademy.app or call/WhatsApp +91 87 5678 7444.