Everything ArthurAI does, counted from the source.

This is the whole system in one place. Nine learning surfaces, a compound reasoning pipeline, live tutoring, five-format assessment, institution administration, reporting, credentialing, and ArthurAI™ Studio — all of it delivered by the Arthur Digital Employee on Eve-Education™ F5/reasoner, with the educator as decider at every layer. Fourteen groups, one system — the full inventory is below, and published in full for machines so an AI crawler reads the same numbers you do.

Nine ways to deliver a lesson — selected per learner, not per course.

• —Nine instructional delivery surfaces: audio, dialogue, microlearning, practice lab, project lab, self-paced, visual, pre-assessment, and collaborative
• —Surface selection is driven by the learner’s cognitive profile and current state — the same lesson can arrive as a dialogue for one learner and a practice lab for another
• —Six-step lesson anatomy underneath every surface: introduction → key concepts → detailed explanation → practice questions → real-world applications → summary
• —Fourteen interactive content block types assembled per lesson — text, callout, definition, diagrams, hierarchy trees, comparisons, tables, timelines, inline quizzes, and more

Perceive, reason, generate — a pipeline, not a single model call.

• —Eve-Education™ F5/reasoner — a compound, multi-stage reasoning architecture, not one model behind a prompt
• —Perceive: every learner message is read before anything is written — question intent, actual comprehension, emotional state, and whether a misconception is forming and how serious it is
• —Reason: a deliberate teaching decision across eight dimensions — strategy, density, modality, tone, emotional response, metacognition, transparency — with the why recorded
• —Generate: the response streams token by token, grounded in the lesson and its sources
• —Calibrated confidence on every output — low-confidence judgments are flagged for a human, not asserted as fact
• —Escalation by design: a stuck learner or a safety gate hands off to the responsible educator with a briefing — the loop never ends at the machine

Six ways to teach a concept — chosen and changed mid-session.

• —Concept-first, example-first, Socratic, scaffolded, problem-based, and analogical instruction
• —Strategy is selected per turn, fitted to the learner and the concept — and Arthur explains its own choice rather than defaulting to one mode
• —Strategies adapt mid-session as the learner’s comprehension and emotional state shift

Live strategy re-selection, with a ledger of why.

• —LENS re-selects the teaching strategy live from learner state — it does not wait for the next lesson to adapt
• —Information density is tuned to the moment: one idea at a time when a learner is struggling, more when they are moving quickly
• —Emotional response handling — frustration or anxiety triggers a change of approach, not more content through a closing window of attention
• —A decision ledger records each adaptation and the reason it was made

The learner is measured before the first lesson — and re-read after every one.

• —The Learner Cognitive Profile (LCP) intake instrument — 30 questions across five fixed categories, measured against four canonical learning-style dimensions
• —The PRISM 25-dimension learner profile derived from ongoing learning behavior — the intake instrument and the derived profile are distinct, by design
• —The cognitive radar — a visual read of the learner’s profile for the learner and their educator

A tutor that knows the lesson, cites its sources, and knows when to refuse.

• —Streaming, hyper-personalized tutoring — learner name, age, language, geography, and current lesson assembled at every turn
• —Selection-to-tutor: highlight any text in a lesson and ask Arthur about it in context
• —RAG-grounded answers anchored to the curriculum and its sources — grounded reasoning, not free-floating generation
• —Saved explanations the learner can return to
• —Explicit refusal boundaries: an educational-intent gate declines non-academic procedural requests, and tutoring is disabled during practice so it cannot write the answer for the learner

Five question formats, engineered against gaming.

• —Multiple-choice, multi-select, true/false, fill-blank, and matching formats on every practice set
• —Anti-gaming rules in generation: options of near-identical length, misconception-probing distractors, the correct answer never consistently the longest
• —Every question carries a 60–120-word explanation — why the right answer is right and why each distractor is wrong
• —Server-authoritative grading — the client cannot manipulate scoring
• —Spaced review, with practice accuracy feeding the next day’s generation as the mastery signal
• —The educator attests every grade before it enters the academic record

The learner’s whole study state, on one console.

• —A dashboard of nine learner-facing widgets over the Arthur console — progress, activity, and study state at a glance
• —Word-karaoke lesson narration and a floating audio control with 0.75×–2× playback speed
• —Per-learner preferences (voice, speed, auto-play) persisted and audited

Built for the way institutions actually run.

• —Courses, students, teachers, and packs — full lifecycle administration
• —Sub-admin delegation with permission-scoped access (teachers, students, courses, analytics, announcements, billing, settings)
• —License-aware bulk roster upload — CSV students, teachers, and full course definitions; over-quota imports return a clear over-by count with opt-in partial acceptance
• —Certificates, analytics, and announcements at institution scope
• —Billing and license-quota tracking per institution, with reconciliation to fix drift
• —Integration keys for the systems the institution already runs
• —Faculty-of-record workflows for accreditation-sensitive deployments

Procurement-grade reporting, out of the box.

• —22 prebuilt reports across seven categories: student, course, pack, teacher, AI usage, license, financial
• —A custom report builder over six data sources with column picking, filters, sort, and grouping
• —Scheduled exports — daily, weekly, or monthly — delivered by email with secured download links
• —CSV and PDF output for every report
• —At-risk learner detection: 7+ days inactive under 30% progress, surfaced on the institution dashboard
• —AI-adoption rate, weekly learning-velocity heatmaps, and per-course performance ranking

Credentials an outsider can verify.

• —A visual certificate designer — institutions compose their own templates on a canvas with placeholder tokens, versioned immutably
• —Pack-based issuance rules with configurable completion thresholds and retroactive issuance for learners who already qualify
• —Public certificate verification — every certificate resolves on a no-login verification page with partial-name privacy and revocation support

It works the roster while the institution is closed.

• —Learner nudges and spaced-review scheduling
• —Session-end processing and learner-profile recalibration
• —Nightly analytics refresh
• —Course ingestion pipelines
• —Escalation monitoring — the human-in-the-loop queue is watched, not just written to
• —Webhook dispatch to institutional systems
• —Data export, erasure, and retention workers behind the data-rights commitments in the Trust Center

Access is a capability, not an afterthought.

• —An accessibility quick-menu on the learning surface
• —Word-karaoke text-to-speech — neural narration with word-level timing for synchronized read-along highlighting
• —Native multilingual delivery, including right-to-left languages — the tutor responds and the curriculum generates in the learner’s language
• —WCAG 2.1 AA posture, documented in the accessibility conformance report

A course-design Digital Employee you direct — and watch think.

• —An authoring agent that drafts, structures, and standards-aligns a complete course from a single brief — with glass-box reasoning you can watch while it works
• —Arthur sits in the workspace alongside it: brainstorm an outline, have it read your own courses before it advises, distil the conversation into a brief, and it remembers what you worked on together
• —48 authoring pages over 283 endpoints — a full course-design surface, not a prompt box
• —Grounded authoring from your own sources, with outcomes mapped, credentialed, and sequenced
• —Compliance Autopilot — a compliance-training track that trains and recertifies a workforce, where a cited quote that is not in the passage it cites is dropped rather than printed
• —Ships anywhere: to an ArthurAI™ edition, to the LMS you already run via SCORM, xAPI, or LTI 1.3, or as verifiable credentials
• —19 of Studio’s capabilities are AI/agentic — itemised, anchored to code, and counted inside the 35+ headline figure, never double-counted

The same substrate, configured per edition.

Every group above ships from one platform. Each edition configures it for its audience — see the edition’s own capability page for the inventory in that context.

• ArthurAI™ SLEGenerally available
• ArthurAI™ ULEGenerally available
• ArthurAI™ VLEGenerally available
• ArthurAI™ CLEGenerally available
• ArthurAI™ FLEGenerally available
• ArthurAI™ PLEGenerally available
• ArthurAI™ StudioCourse authoring & compliance training