# Honen: Complete Platform Reference > **This file is every page of Honen's documentation combined into one document.** > Nothing is omitted and no further fetches are needed. It contains the index page > () followed by all 17 topic documents from > in reading order. > > Honen is teaching and learning infrastructure built by StudyFetch, Inc. It replaces > the learning management system, the instructional design work of building and > maintaining courses, and the content delivery that turns source material into usable > formats. Product: . Contact: support@honen.com > > Contents, in order: > 1. Index and Overview > 2. Platform Overview > 3. The Three AI Surfaces > 4. The Learn Engine > 5. Who Honen Is For > 6. Course Creation > 7. The Learner Experience > 8. Activity and Project Types > 9. Assignments > 10. Sites and Distribution > 11. LMS Integrations > 12. Analytics and Reporting > 13. Funder and Compliance Reporting > 14. People, Groups and Workspaces > 15. Knowledge Base and Parallel Training > 16. Agents > 17. Integrations and MCP > 18. Security and Compliance --- --- # Honen: Index and Overview > Honen is teaching and learning infrastructure for anyone, built by StudyFetch, Inc. "For anyone" is literal: anyone who teaches — a solo expert, a training team, a school, an entire enterprise — can turn their material into a real course without instructional designers or an LMS project, and anyone who learns gets a 1:1 AI tutor rather than a static playlist of videos. Upload documents, manuals, textbooks, or syllabi, and the platform generates reading, video, flashcards, podcasts, quizzes, lectures, arcade games, music-based learning, comics, and AI chat — all from one upload. Every learner gets a 1:1 AI tutor that adapts to how they learn best. ## Why We Exist The world is changing faster than any curriculum can keep up. AI created entirely new career paths in months, but schools take 3–5 years to update a single curriculum — and districts rotate through major reviews only every 5–10 years. By the time a textbook is printed, the skills it teaches may already be outdated. Meanwhile, the systems meant to help people learn are failing. Traditional online courses have 5–15% completion rates. Lecture-based formats don't match how most people actually retain information. And while 47% of students already use AI chatbots for studying, tools like ChatGPT answer questions but don't teach — there's no structure, no progress tracking, no real path to mastery. Access is the other half of the problem. High-quality learning has historically been locked behind expensive institutions, geography, and time. A student in rural Appalachia, a single parent retraining for a new career, and a factory worker learning AI skills all deserve the same quality of education as someone at a top university. Most of them will never get it through the traditional system. Honen exists to close that gap. We believe everyone deserves a personal tutor, a curriculum that updates as fast as the world does, and learning formats that actually work for their brain — regardless of where they live, what they can afford, or how they learn best. When 63% of Gen Z worry AI will eliminate their jobs, the answer isn't to wait for institutions to catch up. It's to give every person the tools to learn anything, right now. ## What We Do Honen turns expertise into structured, multi-format courses in minutes. Instructors upload their materials — PDFs, slides, manuals, syllabi — and AI transforms them into nine distinct learning formats. Every course includes an always-available AI tutor that uses Socratic questioning, identifies weak spots, and adapts to each learner's retention style. The platform serves individual learners, K–12 districts, higher education institutions, enterprises, small businesses, government agencies, workforce development programs, trades and unions, and assessment providers. ### Learning Formats Every course is available in nine formats. Progress carries across all of them: - **Read**: Generated reading content with images, code, and math - **Flashcards**: Term and definition cards with spaced repetition - **Quiz**: Practice assessments with confidence tracking and explanations - **Visual Lecture**: AI-generated illustrated slide decks with speaker notes - **Podcast**: Two-speaker podcast audio with a synced transcript - **Explainer Video**: Generated video with a transcript panel and captions - **Jam Session**: Music-based learning (e.g., concepts turned into songs) - **Arcade Game**: Arcade-style educational games fed by the topic's questions - **Comic**: Visual storytelling with illustrated panels The AI tutor is available alongside all nine rather than being one of them. ### Hands-on Projects Ten project types ask for real work rather than recall: scenario roleplay, agent collaboration, prompt challenge, code notebook, full sandbox, screen share, tagalong (phone camera, for physical skills), interactive walkthrough, 3D simulation, and rubric-graded worksheet. ### Key Capabilities - Upload any document and get a full multi-format course in minutes - 1:1 AI tutor available 24/7 that adapts to each learner and can drive the interface - Course updates add new material without resetting learner progress - Verifiable certificates with unique IDs (shareable on LinkedIn, downloadable as PDF) - Browser-based coding exercises, sandboxes, and hands-on activities - Assign a whole course or any slice of one, with a completion bar per group - Deliver into an existing LMS: LTI 1.3, SCORM 1.2 and 2004, Google Classroom, xAPI and cmi5 - Custom AI Agents on Slack, Microsoft Teams, iMessage, and your own website - Bulk roster imports for organizations (instructors, classes, students) - Cross-format progress tracking and learning analytics - Funder and regulator report compilation (PIRL, NRS, RAPIDS, Perkins V, ETP, ETPL) - Complete per-workspace data isolation ## Platform Documentation Comprehensive reference on what the platform does, written for both people and AI agents. Each document is plain markdown. Everything is also concatenated into for a single fetch. **Start here** - [Platform Overview](https://honen.com/llms/platform-overview.md): What Honen is, the three things it replaces, the three product pillars, and what it explicitly does not do. - [The Three AI Surfaces](https://honen.com/llms/ai-surfaces.md): The Course-Editor agent for content, the Workspace Assistant for operations, and the Widget Builder for custom reporting described in a sentence. - [The Learn Engine](https://honen.com/llms/learn-engine.md): How courses improve themselves from cohort signal without ever resetting learner progress. - [Who Honen Is For](https://honen.com/llms/audiences.md): Enterprise, small business, workforce development, government, higher education, K-12, trades and unions, and assessment providers. **Building and learning** - [Course Creation](https://honen.com/llms/course-creation.md): Every source you can build from, the conversational build process, outline approval, the editor, and versioning. - [The Learner Experience](https://honen.com/llms/learner-experience.md): Format choice, the AI tutor and what it can act on, personalization, citations, assessment integrity, and how progress is actually measured. - [Activity and Project Types](https://honen.com/llms/activities-and-projects.md): All nine activity types and all ten project types, plus how tests work. **Delivering and measuring** - [Assignments](https://honen.com/llms/assignments.md): Assigning part of a course, completion bars per group, self-maintaining cohorts, delivery channels, and LMS grading. - [Sites and Distribution](https://honen.com/llms/sites-and-distribution.md): Share links, catalogs, public storefronts on your own domain, and standalone activity links. - [LMS Integrations](https://honen.com/llms/lms-integrations.md): LTI 1.3, SCORM 1.2 and 2004, Google Classroom, and xAPI and cmi5, how to choose, and running several at once. - [Analytics and Reporting](https://honen.com/llms/analytics-and-reporting.md): Comprehension analytics, conversational reporting, and custom auto-refreshing dashboards. - [Funder and Compliance Reporting](https://honen.com/llms/funder-and-compliance-reporting.md): PIRL, NRS, RAPIDS, Perkins V, ETP and WIOA-aligned outcomes, what Honen auto-fills and what it cannot. **Organization and platform** - [People, Groups and Workspaces](https://honen.com/llms/people-and-workspaces.md): Rostering, roles, nested groups, inactive members, curricula and certificates, terminology, and branding. - [Knowledge Base and Parallel Training](https://honen.com/llms/knowledge-base.md): One approved corpus that trains people as courses and grounds your AI tools as skills. - [Agents](https://honen.com/llms/agents.md): Custom AI assistants on Slack, Microsoft Teams, iMessage and your own website, and how permissions are enforced independently of the prompt. - [Integrations and MCP](https://honen.com/llms/integrations-and-mcp.md): Sources in, Honen's AI into your tools, and Honen's assistants into your channels. - [Security and Compliance](https://honen.com/llms/security-and-compliance.md): Certifications, data handling, identity, Agent safeguards, and assessment integrity. ## What We Do Not Do - Honen does not sell or share learner data - Honen is not a generic AI chatbot — it provides structured courses with curriculum, progress tracking, and verified outcomes - Honen does not require instructional designers or additional trainers to create courses - Honen does not reset learner progress when courses are updated - Honen does not require you to migrate off your existing LMS in order to use it - Honen is not an xAPI or SCORM content player: it produces packages and statements, it does not import and play third-party ones - Honen does not claim to file your funder reports for you. It compiles what it owns and flags what still needs an intake upload or a state wage match ## Services - [For Learners](https://honen.com/learners): Browse courses or upload your own materials. Learn in every format — reading, video, flashcards, podcasts, games, comics, and more. AI tutor included. - [For Enterprise](https://honen.com/enterprise): Turn organizational expertise into training. Upload materials, deploy multi-format courses, track comprehension across teams. - [For Workforce Development](https://honen.com/workforce): Transform manuals and SOPs into workforce training programs. Get more people to competency, faster. - [For Small Business](https://honen.com/small-business): Train your team without a training department. SOPs, onboarding docs, and product guides become courses with an AI tutor. - [For Government](https://honen.com/government): Mission-ready training for government agencies. WIOA-aligned, FedRAMP-ready. - [For Higher Education](https://honen.com/higher-ed): Turn curricula into AI-personalized learning. Faculty retain control. LMS integration supported. - [For K–12](https://honen.com/k12): AI study tools that meet students where they learn. Standards-aligned, FERPA and COPPA compliant. - [For Trades & Unions](https://honen.com/union): Trade-ready training from your own standards. Codebooks, trade manuals, and certifications become multi-format training. - [For Assessment Providers](https://honen.com/testing): Turn assessments into structured prep training from content guidelines and exam blueprints. - [Course Gallery](https://honen.com/browse/gallery): Browse published courses across AI, data science, skilled trades, healthcare, business, software development, and more. ## Key Information - [Homepage](https://honen.com/): Main marketing page with platform overview, features, and partner highlights. - [Accelerate Initiative](https://honen.com/accelerate): StudyFetch's workforce initiative to prepare every American for AI-created jobs, covering K–12 through career pathways. - [Terms of Service](https://honen.com/terms): Legal terms governing platform usage. - [Privacy Policy](https://honen.com/privacy): Data collection, usage, CCPA compliance, and privacy practices. - [Support](https://honen.com/support): Contact support at support@honen.com. ## Security & Compliance - SOC 2 Type II certified - FERPA compliant - COPPA compliant (K–12) - AWS Govcloud (government) - AES-256 encryption at rest - SAML 2.0 / OIDC single sign-on - Complete per-workspace data isolation - Full audit logging ## Partners - **NVIDIA**: AI workforce partner. First-ever adaptation of Deep Learning Institute (DLI) curriculum for K–12 classrooms. - **The College Board**: Strategic investor in StudyFetch, exploring AI-powered college and career readiness. - **Palm Beach County Schools**: MOU with the 10th largest U.S. school district for workforce-ready learning, reaching up to 40,000 high school students. - **Miles College**: HBCU access initiative providing up to 750 students with AI-powered learning tools. ## Contact - Email: support@honen.com - Sales Demo: https://cal.com/team/honen/honen-intro-call - Website: https://honen.com --- # Honen: Platform Overview Honen is teaching and learning infrastructure built by StudyFetch, Inc. It replaces three things organizations normally buy separately: 1. **The learning management system** that hosts and tracks training. 2. **The instructional design work** of building and maintaining courses, whether that is an L&D team, a training department, a curriculum committee, or a courseware vendor. 3. **The content delivery system** that turns source material into usable formats. You supply material you already own. Honen produces a structured course with an always-available AI tutor, delivers it to the people who need it, and reports on whether they actually learned. It can also be added *alongside* an existing LMS rather than replacing it, which is usually the faster path for an organization with an established estate. See [LMS integrations](https://honen.com/llms/lms-integrations.md). ## The three pillars ### A. Practical learning, not videos and quizzes Learners do the work. Ten kinds of hands-on project put them in a real scenario: role -playing a difficult conversation with AI characters, working in a live coding sandbox with a file tree and terminal, being coached through a real tool while sharing their screen, holding a phone camera over physical equipment, or exploring a 3D simulation. Passive formats exist and are useful, but applied practice is the primary modality. Assessment follows from that. Mastery is graded across quizzes, projects and simulations rather than inferred from time spent, and every answer carries a confidence check, which distinguishes a learner who knows the material from one who guessed and one who is confidently wrong. ### B. A course agent instead of an authoring queue Every course has an agent you direct in plain language. Describe a topic or attach files and it proposes an outline, waits for approval, then builds the course: units, topics, readings, quizzes, lectures, projects and tests. Afterwards the same conversation updates, extends, adapts, retimes or translates it. This is what removes the bottleneck. There is no blank editor to learn, no instructional designer to brief, and no queue between deciding training is needed and having it. ### C. The Learn Engine: courses that improve themselves Signal from real cohorts feeds back into the course. Where learners stall, where a question is systematically misunderstood, where a format is ignored, the course is improved so the next cohort does not hit the same wall, and learner progress is never reset by an update. See [the Learn Engine](https://honen.com/llms/learn-engine.md). ## What a learner gets One topic can be taken several ways, and progress counts identically whichever is chosen: read it, watch it, listen to it as a podcast, work through flashcards, read it as a comic, learn it as a song, play it as an arcade game, or be quizzed by a tutor. Two formats can run side by side. An AI tutor is present on every screen, available 24/7 by voice or text, and it can act on the interface rather than only talking about it. During tests its search and generation tools are withdrawn. See [the learner experience](https://honen.com/llms/learner-experience.md). ## What an administrator gets Three AI surfaces, each with its own job: the Course-Editor agent for content, the Workspace Assistant for operations and analytics in plain language, and the Widget Builder for custom reporting described in a sentence. See [the three AI surfaces](https://honen.com/llms/ai-surfaces.md). Around them: assignments that can cover a whole course or any slice of one with a completion bar per group, groups that nest as deeply as the organization does, comprehension analytics with the assigned cohort as the denominator, and report compilation for the frameworks a publicly funded program actually files. ## Who it is for Individual learners, K-12 districts, higher education institutions, enterprises, small businesses, government agencies, workforce development programs, trades and unions, and assessment providers. The product is the same in each case; the vocabulary, source material and reporting frameworks change. See [who Honen is for](https://honen.com/llms/audiences.md). ## What Honen does not do - It does not sell or share learner data, and enterprise data is not used to train AI models. - It is not a general-purpose chatbot. Courses have structure, assessment, progress tracking and verifiable outcomes. - It does not require instructional designers or a training department. - It does not reset learner progress when a course is updated. - It does not require you to migrate off your existing LMS in order to use it. - It is not a SCORM or xAPI content player. It produces packages and statements; it does not import and play third-party ones. - It does not claim to file funder reports on your behalf. It compiles what it owns and flags what still needs an intake upload or a state wage match. - It is not an authoring tool you have to learn. The primary interface is conversation. ## Reference documents **Core** - [The three AI surfaces](https://honen.com/llms/ai-surfaces.md) - [The Learn Engine](https://honen.com/llms/learn-engine.md) - [Who Honen is for](https://honen.com/llms/audiences.md) **Building and learning** - [Course creation](https://honen.com/llms/course-creation.md) - [The learner experience](https://honen.com/llms/learner-experience.md) - [Activity and project types](https://honen.com/llms/activities-and-projects.md) **Delivering and measuring** - [Assignments](https://honen.com/llms/assignments.md) - [Sites and distribution](https://honen.com/llms/sites-and-distribution.md) - [LMS integrations](https://honen.com/llms/lms-integrations.md) - [Analytics and reporting](https://honen.com/llms/analytics-and-reporting.md) - [Funder and compliance reporting](https://honen.com/llms/funder-and-compliance-reporting.md) **Organization and platform** - [People, groups and workspaces](https://honen.com/llms/people-and-workspaces.md) - [Knowledge base and parallel training](https://honen.com/llms/knowledge-base.md) - [Agents](https://honen.com/llms/agents.md) - [Integrations and MCP](https://honen.com/llms/integrations-and-mcp.md) - [Security and compliance](https://honen.com/llms/security-and-compliance.md) --- # The Three AI Surfaces Honen is not one chatbot with different labels. Three distinct AI surfaces do three different jobs, each with its own permissions and its own place in the product. ## 1. The Course-Editor agent One agent per course, directed in plain language. It stands in for the instructional-design work of building and maintaining a course. It can author every activity type and every project type, generate and place images, produce and edit explainer videos, import PDF pages as lecture slides, edit walkthroughs down to individual hotspots, crawl and cite sources, translate a whole course, and restructure to a time budget. It validates its work and compiles back to the live course. Representative instructions: - "Build a new-hire onboarding course from this handbook." - "Add the latest code changes to every course in this program." - "Translate the entire safety course into Spanish and Portuguese." - "Tighten every module to a 20-minute time-on-task budget." - "Rewrite the lessons for first-year apprentices." - "Build a 14-week course from this textbook and syllabus." - "Convert the lectures into 10-minute audio podcasts." Everything it produces is reviewable before learners see it, changes accumulate in a draft, and every applied change is a revision with rollback. It handles production; a person keeps the editorial and academic decisions. See [course creation](https://honen.com/llms/course-creation.md). ## 2. The Workspace Assistant The operations surface. The same chat that answers analytics questions also runs the workspace, which means administration does not require a tour of an admin console. **It reads:** engagement, mastery, time-on-task, certification status, at-risk learners, per-team, per-cohort, per-site, per-group rollups, and any individual learner's full picture. **It writes:** bulk-invite a cohort from a list of emails or a CSV, create and nest groups, assign a course or a slice of one with a due date and inclusion or exclusion lists, promote someone to instructor, mark a member inactive without losing their progress, and retire a cohort. Representative instructions: - "Bulk-invite this intake by email and drop them in the Field Crew group." - "Assign Q4 compliance training to all 1,200 employees, due December 15." - "Spin up a Sales Enablement / EMEA group and seed it with the reps in scope." - "Mark this worker inactive while they are laid off, without losing progress." - "Who is past their 30-day onboarding due date?" Destructive and bulk operations render a confirmation bar before anything happens, and the agent cannot fire the operation until the person clicks through it. That covers removing a member, deleting a group, any workspace-wide assignment, five or more invitations at once, three or more role changes, and changes to workspace settings. **Hand-off to the editor.** Any "fix this" request becomes a one-click card that opens the exact place in the Course-Editor agent. The Assistant finds the problem; the Editor fixes it. Course inspection works at four levels of zoom, from the list of editable courses, to course structure, to topic detail, to the body of a single activity, each with a direct deep link. Reads that expose roster-wide data require management permission, so a learner asking on a chat channel cannot pull the cohort's numbers. It uses the same code paths as the dashboard, so an assistant-created assignment is indistinguishable from a hand-made one. ## 3. The Widget Builder Custom reporting, described in a sentence. Every workspace ships a standard analytics dashboard; the Widget Builder is for the question the standard dashboard does not answer. Describe the cut you want and it becomes a permanent, auto-refreshing dashboard tile. No report builder, no analyst queue, no data-team ticket. Refresh runs unattended on an interval you pick, from five minutes up to daily, so the cadence matches the decision: a live cohort during a rollout, hourly aggregate KPIs, a daily board number, a monthly funder submission. A failed refresh keeps the last good render and surfaces the error rather than showing a broken tile. **Read-only by design.** A widget cannot contain buttons that change anything, which is what makes it safe to leave running unattended on a screen in an office. Examples of what one sentence produces: - "Certifications expiring in the next 90 days, grouped by manager." - "Time-to-complete on the new-hire track by start month." - "Compliance pass rate by region, with a manager drill-down." - "Managers whose team has zero completions on the Q4 rollout." - "DFW rates by section." - "At-risk students four weeks before state testing." - "Journeyman-exam pass rate by local by year." - "Pre-apprenticeship to apprenticeship conversion rate by area." - "Outcomes by cohort and zip code." **Deterministic by construction.** It runs the same read-only queries the Workspace Assistant uses, then writes a script that aggregates the result into a chart specification. The same queries return the same shape on every refresh; only the values change. That is what makes the output usable for an audit or a regulator rather than only for internal curiosity, because a number in a report can be traced to how it was produced. ## How they relate The three are designed to hand off to each other. The Workspace Assistant surfaces that a module is failing a cohort; one step takes you into the Course-Editor agent to fix it; the Widget Builder pins the measure that tells you whether the fix worked. And underneath all three, the [Learn Engine](https://honen.com/llms/learn-engine.md) closes the same loop automatically. Separately from these three, a workspace can build its own [Agents](https://honen.com/llms/agents.md) for its people to talk to on Slack, Microsoft Teams, iMessage, or an embedded chat on its own website. --- # The Learn Engine Courses normally decay. They are written once, deployed, and then slowly drift away from what learners actually need, until someone schedules a review cycle. The Learn Engine is the part of Honen that closes that loop automatically. It is closed-loop auto-improvement: courses are self-improving, so each cohort lands on a better version than the one before it, with no review meeting in between. It watches signal across cohorts, identifies where learners stall, and feeds that back into the course, so each cohort lands on a better version than the last. ## The loop 1. **Telemetry.** Every activity reports real engagement rather than a completion click: how far into a reading someone got, how long they spent, which quiz questions they missed, how confident they said they were, which format they chose, where they abandoned. 2. **Detection.** Patterns are identified across the cohort rather than per person: a topic where most learners drop, a question almost everyone gets wrong, a format nobody picks, a unit where time-on-task is triple the estimate. 3. **Intervention.** For the individual learner, the AI tutor switches format, queues reinforcement, or changes approach in the moment. 4. **Improvement.** For the course itself, the improvement is applied so the next cohort does not hit the same wall. ## Progress is never reset This is the constraint that makes automatic improvement safe to switch on, and it applies to every kind of course update. When a course changes, learners keep the ground they have earned. New or altered material appears as something to review, not as lost progress, and a learner who had completed a unit does not find themselves incomplete because the unit was improved. Nobody is punished for having taken the course early. For a compliance or regulated context this matters twice over, because the audit trail survives the update as well: revisions are recorded with author and timestamp, and any revision can be previewed or rolled back. ## Confidence, not just correctness Every quiz and test answer is followed by a confidence check. That produces a signal a score alone cannot: the difference between a learner who knows the material, one who guessed correctly, and one who is confidently wrong. The last group is the one traditional analytics miss entirely, and the one most worth intervening on. ## What this replaces A curriculum review cycle. Many programs take three to five years to update a curriculum, and some institutional review cycles run longer, while the skills being taught change in months. The Learn Engine is the argument that a course can improve continuously from evidence instead of waiting for a committee, and that the evidence is the cohort's own behaviour rather than a survey. Manual improvement remains available at any time through the [Course-Editor agent](https://honen.com/llms/ai-surfaces.md), driven by the same analytics. --- # Who Honen Is For The platform is the same for every audience. What changes is the vocabulary, the source material, the regulatory framework, and which reports matter. Every workspace can rename what its users see, so groups become cohorts, sections, locals or departments, and members become students, employees, apprentices or crew. Honen consolidates three things in every case: the learning management system, the work of building and maintaining courses, and the content delivery that turns source material into usable formats. ## Enterprise Large L&D, talent and HR functions running big catalogs across many teams and geographies. Chat your catalog updates: build onboarding from the handbook, refresh compliance training against new regulatory guidance, translate a safety course into several languages, tighten every module to a 20-minute time-on-task budget. Everything is reviewable, and no instructional designer sits in the middle. The operations surface handles the scale: bulk-invite a cohort from a CSV, assign Q4 compliance to twelve hundred people with a due date, create a regional group and seed it, promote a senior individual contributor to instructor, retire last year's cohort. Custom dashboards give leadership what it reads in standups: certifications expiring in 90 days grouped by manager, time-to-complete on the new-hire track by start month, compliance pass rate by region with a manager drill-down, managers whose team has zero completions on a rollout. Primary use cases are onboarding, upskilling and compliance, with board-ready outcome reports covering engagement, mastery, time-on-task and certification across every team and program. Hands-on AI projects (spreadsheets, image generation, research, web pages, role-play) address AI-readiness directly. ## Small business Companies with no dedicated L&D function: owners, operations leads and people-managers wearing the training hat. The relevant comparison is what a training capability normally costs, which is an enterprise LMS plus instructional design plus content vendors, and often runs into tens of thousands of dollars a year before anyone has been trained. Honen scales from five to five hundred employees without that stack. The course agent *is* the L&D function: "build a 30-60-90 onboarding plan from these SOPs" or "update onboarding with this revised handbook" returns a complete multi-format course. The assistant is the whole HR and training console in plain English: invite a new hire, drop them in a group, assign onboarding with a 30-day due date, promote the lead trainer. Pin one or two reports and never log in deeper than that. Setup is measured in days. Upload SOPs Monday, assign onboarding Friday, with no RFP and no implementation partner. Typical content: 30-60-90 onboarding, product training, customer-support certification, release notes, sales playbooks. ## Workforce development Workforce boards, sector partnerships, registered apprenticeship sponsors and state agencies running pre-apprenticeship and upskilling programs. The problem is timing. Programs often take three to five years to update a curriculum while the skills expire in months. Subject-matter experts (master journeyworkers, lead trainers) direct the agent themselves: add the latest code changes to every course, build a six-week pre-apprenticeship program from these intro materials, convert the lectures into ten-minute podcasts for the truck. New training deploys the same day. Program operations run per cohort, per shift and per local, including marking a worker inactive during a layoff without losing progress. Reporting is built for the funder: outcomes by cohort and area, time-to-certification by trade, completion ratios by quarter, on a refresh cadence that matches the reporting cycle. See [funder and compliance reporting](https://honen.com/llms/funder-and-compliance-reporting.md). Apprentices run hands-on simulations of the actual job, applying real standards in realistic situations, and course updates never reset worker progress, which matters when codes shift mid-cohort. ## Government Federal, state and local agencies running both internal compliance training and outward-facing community workforce programs. Compliance is architectural: SOC 2 Type II, a FedRAMP-ready posture, FERPA, full audit trails, and per-agency and per-program data isolation, with AWS GovCloud (US) deployment available. Outcomes map to the federal frameworks agencies report against. The agent replaces the vendor and procurement cycle: build an awareness course from updated guidance, refresh compliance training with this year's policy changes, roll out the update without resetting employee progress or audit trails. Reporting is pinned per regulator and cadence, read-only and audit-clean, with each refresh running a deterministic script the agency can review. Rollup insight surfaces without exposing individual raw data, which is the distinction between workforce visibility and workforce surveillance. It is a two-sided deployment: internal employee compliance and upskilling, plus community-facing programs such as AI literacy and career pathways, on one platform. ## Higher education Community colleges, four-year institutions and consortiums; provosts, deans, faculty and courseware buyers. The economic argument is against licensed courseware and publisher access codes, which charge per learner or per course for content an institution does not own. Honen generates courses from materials the institution already has. Faculty control the agent and review every output before students see it: build a 14-week course from this textbook and syllabus, add the latest journal publications, rewrite the lessons for first-generation college students. The agent handles production, not academic decisions. Departmental operations cover rostering from a registrar export, nested section groups, assigning a syllabus with a due date, and dropping a tutoring add-on as a scoped assignment for at-risk students. Institutional-research questions faculty and deans can build themselves: DFW rates by section, time-to-mastery on a gateway course by major, completion lift after a redesign, at-risk students four weeks before midterms. It fits the existing campus stack rather than demanding replacement, over LTI 1.3, SAML and CAS single sign-on, and partner-built courses can be deployed inside the same student experience as faculty-built coursework. Assessment is exportable for accreditation and institutional reporting. ## K-12 Superintendents, curriculum directors, district IT and classroom teachers. Safety positioning is the strongest here: FERPA compliant, COPPA aligned, age-appropriate AI guardrails, student data never sold or shared, single sign-on through Clever and ClassLink, and per-district data isolation. Teachers and curriculum directors direct the agent: build a course aligned to our ninth-grade standards, rewrite the lessons for advanced learners, translate the entire course into Spanish for our newcomers, tighten every module to fit a 30-minute class period. Generation works against state standards and exam blueprints, and rewrites for grade level on demand. Test-prep covers practice tests, study plans and weak-area drills. District operations nest cleanly from district to school to classroom period, including a translated variant as a scoped rollout for newcomer students. Reporting covers every layer: mastery by grade by school, weekly progress by classroom, newcomer growth in each language version, at-risk students four weeks before state testing, refreshed hourly during testing windows. Ready-made tracks on AI literacy, career exploration, coding and financial literacy deploy alongside district curriculum. The classroom dynamic stays human: teachers guide while the AI handles per-student tutoring, which is what frees teacher time. ## Trades and unions Building trades unions, trade associations, apprenticeship sponsors and joint apprenticeship training committees. This is a full-pipeline deployment: interesting high-school and college students in the trades, then pre-apprenticeship, apprenticeship, journeyworker upskilling, and license-exam preparation. The subject-matter experts of the trade direct the agent: spin up a journeyman certification track from the codebook, update the course with the latest code changes, rewrite the lessons for fifth-year apprentices ready for the exam. Source material is trade-specific: electrical codebooks, safety manuals, plumbing standards, welding procedures, HVAC specifications. New training reaches every local in minutes. Program operations run per local, per cohort and per certification level, including marking a member inactive during a layoff without losing progress, and issuing a translated update only to the locals that asked for it. Reporting is per stakeholder: exam pass rate by local by year, update completion by crew, pre-apprenticeship to apprenticeship conversion by area, simulation completions during the busy season. Apprentices run troubleshooting scenarios, code application and realistic installations. Delivery is tuned to the work: translated for crews, and lectures converted to ten-minute podcasts for the drive home. ## Assessment and testing providers Organizations whose product is the assessment itself use the same authoring and delivery, with per-tenant isolation, item-level analytics, confidence tracking alongside correctness, and packages delivered into a client's own LMS with separate signed credentials per client. --- # Course Creation Most tools in this category give you a blank editor. Honen starts from the opposite end: you describe what you want, or attach material you already have, and a course agent builds the course while you watch and redirect it in conversation. ## What you can start from You need either some text or one processed file. You can supply both. | Source | Notes | |---|---| | Files | Any type, up to 1 GB each. PDF, Office, text, images and video are parsed automatically. Other formats become a raw source the agent opens and reads. | | Screen recording | Recorded in the browser. Becomes an interactive walkthrough by default, not a video. | | YouTube | By link, transcribed server-side. No upload needed. | | GitHub repository | Public repositories are downloaded and flattened into one readable source. | | Knowledge base | Documents from your workspace knowledge base, snapshotted into the course. | | Granola | Meeting notes and transcripts, where the connection is configured. | | MCP servers | Linear, Notion, HubSpot, Google Drive and others, available on the agent's first turn. | Attached material is processed and read before the agent proposes anything, so its first response refers to specifics in your document. Status is shown per file, and course creation stays disabled until processing finishes, so the agent never starts against a half-read source. ## Controls at creation - **Brand.** Sets the visual identity for generated imagery: Editorial Doodle, Professional, Playful, or Custom, with optional name, logo, colour overrides and free-text guidelines. Defaults to the workspace brand. - **Research.** When enabled, the agent searches the web and saves research notes before proposing an outline, so units are grounded in current sources. - **Organization skills.** Workspaces can author reusable prompt fragments that inject house style or required framing into the brief. ## How the build works The agent reads first. It crawls any URL you provided, opens every attached source, and searches the knowledge base when one is available. Only then does it ask what it genuinely does not know: who the course is for, their level, how it should be structured, and how long it should be. Anything already implied is skipped. It then proposes a full outline: units, topics, learning goals, activity types per topic, a project per unit where one fits, and a test plan per unit. **Nothing is built until you approve the outline.** Ask for changes in plain language and it revises in place. On approval it creates the structure and generates every activity in parallel. You can leave while it builds. Length ranges from roughly three units of two topics up to six units of four, or a custom shape you name. Structure presets are Academic (a reading, flashcards and a quiz per topic), Project Based (a reading per topic and a project per unit), or Custom. ## The editor Four views share one persistent chat sidebar: - **Preview** is the real learner runtime pointed at your draft, not a mock. You can switch between the author's view and exactly what a learner sees. - **Advanced Edit** is the structural editor: a course tree with panels for each unit, topic, activity, test and project. - **Materials** manages sources after creation. Knowledge-base-linked material raises an alert when the original document changes, with one-click refresh. - **Settings** covers details, marketing copy, people, AI style, brand, tutor, guardrails, citations, visibility and cloning. The agent can generate and place images, produce and edit explainer videos, import PDF pages as lecture slides, edit walkthroughs down to individual hotspots, and cite sources. Image generation covers everything a course needs visually and all of it from the same conversation: course thumbnails, lecture slides, comic panels, podcast cover art and project reference images. Its output is validated rather than trusted: structure is schema-checked, every identifier is inventoried, and an automatic compile step writes the change back to the live course. It confirms before destructive or bulk actions and verifies its own work before reporting it complete. ## Nothing is destructive Changes accumulate in a working draft. Applying a draft creates a revision. Version history lists every revision with author and timestamp, a preview link, and rollback on any of them, and a rollback creates a new draft rather than discarding anything. Every message you send is a checkpoint. Drafts stay private until you publish, and publishing is a separate, deliberate step. ## Keeping courses current Courses are updated the same conversational way they were built, including from what the analytics show: where cohorts stall, which topics learners struggle with, which formats land. Learner progress survives updates. --- # The Learner Experience Honen supports your content and puts every learner on their own optimal path to mastery, with a 24/7 AI tutor alongside them. The result is faster completion, deeper understanding, and people who are actually ready to do the work. The learner chooses the format. Authoring one topic produces several ways through it, and every path counts the same progress. ## Choosing a way in Each topic opens on a start screen: where the learner is, what they will learn, and the ways to take it, grouped into three phases. One option is recommended based on what they have done and what they prefer, and that phase is already open. Where only one option exists, the picker is skipped. | Phase | Options | |---|---| | Learn | Reading, visual lecture, explainer video, podcast | | Practice and reinforce | Flashcards, comic, jam, arcade | | Prove | Quiz | Full detail in [activity and project types](https://honen.com/llms/activities-and-projects.md). **Split view.** On desktop a learner can open a second pane and run two formats side by side: read while the podcast plays, or keep flashcards next to the article. Mobile collapses to a single pane. ## The AI tutor A tutor is available on every activity, by voice or text, and persists as the learner moves between topics. Courses can name it and give it a personality and avatar. What distinguishes it is that it can act on the interface, not just talk: flip a flashcard, shuffle the deck, select a quiz answer and reveal the explanation, mark a reading complete, play or seek the podcast, turn a comic page, change the song, open a second pane. When it wants to move the learner to a different topic it has to ask first. In standard mode it is a conversational tutor with tools matched to the current activity. In advanced mode it uses a stronger model, writes formatted answers with code and images, accepts file attachments, and (only where the course creator has enabled each capability) can search the knowledge base, search the course, generate images, and move the learner on. It notices when a learner stalls, switches format on its own, queues the right flashcards at the right moment, and uses Socratic questioning rather than supplying answers outright. **Assessment integrity.** During a test the tutor's search and image tools are withdrawn, enforced both in the agent and on every request. Attachment reading is disabled whenever the learner has an unsubmitted test anywhere in the course. ## Personalization A learner who has completed onboarding can turn on personalization for a reading, and it rewrites around their interests, level and any instructions they have written, for example "short explanations with examples, no jargon". The rewrite streams in place and progress counts identically to the original. Personalization currently applies to readings. Separately, the same course adapts for different audiences and reading levels, and translates into Spanish and other languages on demand. A module can be tightened to a 30-minute class period or a 10-minute mobile podcast without rebuilding the course. ## Citations in readings Where a reading is grounded in a knowledge base source, claims carry a numbered inline citation. Hovering shows the source title and the cited passage; clicking opens the full source, and workspace members can follow through to the document itself. Visibility is configured per course across six levels: hidden, attribution (name only), excerpt (name and passage), workspace (full source for members, the default), granted (full source for those with knowledge base access), and full (everyone). Authors additionally see a warning when a cited passage no longer matches the current source; learners do not. Readings built purely from an uploaded file are grounded in that file but do not produce citation links. ## Progress and completion Nothing completes on a "next" click. Each activity completes on a real signal: - a reading at 95% scrolled - flashcards at 80% of the deck seen - a video at 90% watched - a quiz on submission - a lecture or comic when every slide or page has been viewed A reading containing interactive widgets only counts once the widgets have been used. A learner who is not finished sees a specific list of what remains, not a percentage. Finishing opens a completion screen with a letter grade, a certificate card and the achievements earned. Consecutive-day streaks run alongside, with milestone emails at 3, 7, 14 and 30 days, and twenty achievement badges across completion, variety, tutor time, mastery and streaks. Certificates are verifiable, carry unique IDs, are shareable to LinkedIn and downloadable as PDF, and are backed by AI-graded mastery across quizzes, projects and simulations rather than time spent. Certificates are issued per curriculum rather than per individual course. ## Standalone activities Any topic or single activity can be delivered on its own through a standalone link: no course navigation, just the work and a single onward action when it is done. Progress still counts toward the course. This is what makes a Honen activity usable inside a chat message, an email, an onboarding flow, or another product. --- # Activity and Project Types A course is built from activities and projects. Activities teach and check understanding; projects ask for work. Authoring one topic can produce several activity types, and progress counts identically whichever a learner chooses. ## The nine activity types | Type | What the learner gets | |---|---| | **Reading** | An article with images, code, mathematics and callouts. Scroll progress on the edge, read time and word count. Can contain interactive widgets and embedded walkthroughs inline. Resumes where they left off. | | **Flashcards** | Flip cards with optional hints, keyboard navigation, shuffle, and a definition-first mode. | | **Quiz** | Multiple choice, true/false, short answer, fill in the blank and free response. Every answer is followed by a confidence check; explanations reveal after answering. | | **Visual lecture** | A swipeable card stack of illustrated slides with speaker notes. | | **Podcast** | A two-speaker audio episode with a synced transcript that highlights along with the audio. | | **Explainer video** | A generated or uploaded video with a transcript panel and optional captions. | | **Comic** | Three pages of four panels, readable as a flat spread or a 3D book. | | **Jam** | A song per learning goal, across genres, with synced lyrics. | | **Arcade** | A WebGL game (endless runner, jetpack quiz, platform jump) fed by the topic's own questions. | The AI tutor is available alongside every one of these. It is not itself one of the nine formats; see [the learner experience](https://honen.com/llms/learner-experience.md). ## The ten project types Projects are where a unit stops testing recall and asks for work. | Project | The learner | |---|---| | **Scenario** | Role-plays with AI characters, in text or voice, scored against criteria. | | **Agent** | Works alongside a live AI agent: image generation, HTML, writing and research, spreadsheets, or a full-stack application deployed to its own live subdomain. | | **Prompt challenge** | Writes prompts to reproduce a target image or website, scored on similarity. | | **Code notebook** | Works through Jupyter-style cells in Python, C++ or CUDA, graded by hidden tests plus a rubric. | | **Full sandbox** | Gets a real file tree, terminal and live preview in a browser IDE. | | **Screen share** | Shares their screen and is coached through a real tool, step by step. | | **Tagalong** | Scans a QR code and uses their phone camera for physical, hands-on skills. | | **Walkthrough** | Clicks through hotspots on annotated screenshots taken from a screen recording. | | **Simulation** | Explores a prebuilt 3D environment. | | **Form** | Fills a generated worksheet, graded against a rubric. | In practice this means learners build spreadsheets, generate images, write research, write code, build web pages, role-play customer conversations, apply electrical code to a wiring problem, walk through a safety procedure with a coach watching their screen, or hold a phone over real equipment. ## Tests Unit tests run in their own view with multiple choice, true/false, short answer and interactive artifact questions, plus a timer, clickable progress indicators and confidence tracking. The pass mark defaults to 80% and can be raised per assignment. Where a test is not required, results are presented as informational rather than pass or fail. --- # Assignments An assignment turns a course from something a learner *could* take into something they *owe*. It names a course, or a slice of one, a set of people, an optional deadline, and what "finished" means for that particular group. The same course can be assigned three times to three teams with three different completion bars and three different slices of content, and every learner is measured only against the bar of the assignment they are on. ## Assigning part of a course An assignment can cover a slice rather than a whole course, with no separate course to build and maintain. | You assign | The learner owes | |---|---| | A unit | Everything in it: every topic, its test, its projects | | A topic | That topic's activities | | A single activity | That activity alone | | A unit test | That test alone | | A project | That project alone | These mix freely, and completion is measured against the narrowed content only. Links take the learner straight to the assigned work rather than a course overview. ## Defining completion Every assignment carries its own completion bar: - **Activities per topic**: one, a specific number, or all of them. - **Minimum quiz score**: a quiz counts toward its topic only at or above the bar, measured on the learner's best attempt. - **Unit tests**: required or ignored, with an optional minimum score. - **Projects**: required, or excluded from the bar for this group. Where several whole-course assignments cover the same learner, the strictest value of each setting applies, so nobody gets an easier bar by being assigned twice. Test bars default to 80%. ## Recipients Recipients are built from a tree: the whole workspace, any group including its subgroups, named individuals, or any combination. Exclusions are first-class, so "everyone in Engineering except these two" is a rule set once rather than a list maintained forever. The cohort maintains itself. Someone who joins Engineering picks up every assignment targeting Engineering, including ones created months earlier; someone who leaves loses them. Individuals can be toggled on or off a single assignment, and the reason a person is included is always visible: directly assigned, via a group, workspace-wide, or excluded. Being assigned a course grants access to it, so there is no separate sharing step. Assignments can also be created through the API, so an event in another system can issue the right training at the right moment, and from inside an LMS through LTI deep linking or the Google Classroom add-on. ## Reaching the learner Delivery is in-app and by email, plus Slack, Microsoft Teams, Google Chat and iMessage. Administrators control which channels push, and individuals can opt out. Channel messages carry a signed auto-login link that lands the learner directly on the assigned work; removing someone from an assignment invalidates their outstanding links. Due-soon and overdue alerts fire automatically on thresholds the administrator sets, once per person, with an optional custom message. Learners can also ask the messaging bot what is due and get their assignments back sorted by date. **Due dates are advisory by design.** A passed date marks the assignment overdue and triggers the alert, but never blocks access, locks content or penalizes a score. ## What the learner sees "My Assignments" collects assigned work with tabs, counts, filters and progress. The what's-left checklist states the bar in plain language and then itemizes every outstanding topic, test and project, with each row opening that specific item. ## What the instructor sees Completion, average score and the bar in plain English across the assignments list; an assigned-to table of every learner grouped by their group with progress, score, time and status, sortable and filterable including overdue only; a drill-down from a course's projects to a project roster to an individual submission rendered per project type; and a queue of submissions awaiting review, oldest first. Reminders are sent per learner and adapt to how far along the person is. The system will not remind someone who has already finished, and reports when a learner will not receive one because they have unsubscribed. ## Reporting Every assignment reports completion across the assigned population, test scores, project performance, time spent per activity, and which units and topics people struggle with overall. Course analytics can be narrowed to one cohort and are measured against that assignment's bar, so a one-unit assignment never reports a fraction of its real completion. Any individual learner can be opened in full. Two lenses exist deliberately and are never merged: a learner sees their progress against the strictest bar covering them, while each instructor sees them against their own assignment's bar. ## LMS grading An assignment is the unit that creates a gradebook column in an external LMS; linking a course without one creates no column. Grades post server-to-server over LTI 1.3, and as a draft grade for the teacher to publish in Google Classroom. You choose what a score means per line item: completion, pass/fail, or progress. A scoped assignment is graded purely on its own completion. ## Managed conversationally The workspace assistant can list assignments, inspect one including every assignee's live progress, show one person's assignments across the workspace with overdue flags, create assignments to any scope with due dates and inclusion or exclusion lists, and remove them. It uses the same code paths as the dashboard. Setting completion criteria and content scope remain in the editor. Reads that expose roster-wide progress require management permission, so a learner cannot pull the cohort's numbers. ## Boundaries - Certificates are issued per curriculum, so finishing a scoped assignment earns credit, analytics and an LMS grade but not a certificate. - There is no CSV export of assignment data. Funder and regulator reporting is a separate capability; see [analytics and reporting](https://honen.com/llms/analytics-and-reporting.md). - There are no bulk actions: no duplicate, no bulk selection, no bulk reminder. - Scoped LMS launches land on the whole course. The scope-aware deep link is used by the messaging channels. - Joining a group later adds the work correctly but sends no notification; those fire on creation and on edits that add people. - Assignments require an organization workspace and are not available in personal workspaces. --- # Sites and Distribution Once a course exists there are several ways to get it to people, and they can all be used at once. | Channel | What it is | |---|---| | **Share link** | One permanent link per course. Reach is set per course: workspace only, request access, or anyone with the link, within a ceiling the workspace administrator sets. | | **Explore** | An in-workspace catalog with curated sections that can target specific groups. Off by default for team workspaces until an administrator curates it. | | **Sites** | Branded public course storefronts on your own domain. Covered in detail below. | | **Assignments** | Names a course or a slice of it, a cohort, an optional due date and its own completion bar. See [assignments](https://honen.com/llms/assignments.md). | | **LMS** | LTI 1.3, SCORM 1.2 and 2004, Google Classroom, and xAPI or cmi5. See [LMS integrations](https://honen.com/llms/lms-integrations.md). | | **Standalone activity links** | A single topic or activity delivered on its own, with progress still counting toward the course. | --- # Honen Sites Honen Sites turns selected courses into a branded public learning destination with its own home page, catalog and full course experience, while authoring, analytics and learner management stay inside the same workspace. **Sites is the public front door. The workspace is the house.** One content pipeline serves internal teams and external community learners, and you decide what each audience sees. ## The problem it solves Organizations that want to offer courses outside their internal LMS usually stitch together separate tools: | The usual approach | What breaks | |---|---| | A marketing site plus a separate course platform | Two systems to maintain, inconsistent branding, learners bouncing between URLs | | A public course library that only links out to enroll | No real preview; visitors cannot experience anything before signing up | | Opening the full internal LMS to the public | Internal roles, permissions and cohort data get messy quickly | | A custom-built branded learning portal | Expensive, slow, and diverges from the real product over time | | A documentation or help site | Static reading only. No activities, no tutor, no evidence anyone learned anything | ## What it replaces - **Course storefront platforms** such as Thinkific, Teachable and Kajabi, when Honen is already your LMS and you do not want a second content system. - **Public course landing pages plus a separate enrollment flow.** The Site *is* the enrollment surface. - **Custom-branded learning portals** built by agencies or internal developers. - **Link-out resource hubs** that only list PDFs and video links. - **Documentation and developer-docs portals** such as Mintlify, ReadMe, Docusaurus and GitBook. Product docs, API references and SDK guides become interactive courseware with real activities and an in-context AI tutor. - **Customer-education and enablement academies** in the Skilljar mould: product training, certifications and community enablement from the same workspace instead of a separate platform and content pipeline. - **Community, alumni and partner access** currently managed through ad-hoc invites and shared links. Sites does not replace your internal workspace. Internal instructors, cohorts, assignments and analytics stay where they are. ## Key benefits **One platform, two audiences.** Build courses once, publish a subset publicly. Internal and external learners share the same underlying content and see only what you expose. **Try before committing.** Visitors browse the catalog, open a course and start learning with no account, and they get the real experience rather than a watered-down demo. One login unlocks the tutor, projects and saved progress. **Your brand, not ours.** Logo, social preview image, custom navigation links, custom footer, your own subdomain on day one, and custom domains such as learn.yourcompany.com. **Clean external isolation.** People who sign up become community learners: a dedicated role with its own permissions and usage limits, landing in a per-site group, seeing only that site's courses, and getting a branded home rather than an admin dashboard. **No duplicate content work.** The public experience uses the same components as internal courses, so what visitors preview is exactly what they get after signing up. **Curated like a storefront.** Arrange courses into rows: "Start here", "Advanced tracks", "New this quarter". Hide, reorder and publish when ready. **SEO-ready by default.** Published sites are indexable, course detail pages carry proper metadata, and you control the social preview image, so a catalog can appear in search rather than sitting behind a login wall. **Controlled costs for public learners.** External learners have their own entitlement limits, configurable per workspace, so you decide how much free access to offer before a paid seat or internal enrollment. ## What is gated, and what is not **Available with no login:** readings, flashcards, quizzes, lectures, podcasts, comics and games, with free navigation through the lesson sidebar. **Requires login:** the AI tutor, hands-on projects, saved progress, and the full course experience. Public browsing is read-only until login, so there are no phantom learner records. After signing in, a visitor is auto-enrolled as a community learner in the site's group, keeps their deep links, and continues where they left off. ## Documentation, developer relations and customer education This is the use case most often missed. Static docs tools stop at reading; academy platforms give you customer training as a separate platform and content pipeline. Sites collapses both into the workspace you already run. Point product docs, API references, SDK guides or onboarding playbooks at the Course-Editor agent and they become interactive courseware: the same reading a docs site would show, plus quizzes, hands-on coding and sandbox activities, and a tutor that answers questions in context. Anyone can browse with no login; one login takes a developer or customer into the tutor, runnable projects and saved progress. - **Developer relations.** Ship API and SDK education on your own domain: public, SEO-indexed, hands-on, and matched to the same source your team maintains. - **Customer education.** Product training, certifications and enablement tracks without a second subscription. - **Community enablement.** Open courses with gated advanced tracks behind signup, and clean isolation from internal operations. ## Setting one up Create a site and name it, drag published courses into sections, configure branding and SEO, and publish. Optionally point your own domain at it. Workspace administrators manage Sites by default, and can grant a Manage Sites permission to instructors who should own the public storefront without full admin access. **A workspace can run several sites at once.** A university might run separate continuing-education and alumni sites with different catalogs; a company might run a partner training portal alongside a community hub. Where a workspace has a primary custom domain configured, internal users signing in can be redirected to the branded domain automatically, so the custom domain acts as a full application alias rather than only a public skin. ## Concepts | Concept | Meaning | |---|---| | Site | A curated, publishable storefront belonging to one workspace | | Site sections | Rows of courses on the home page, like explore categories | | Subdomain | Provisioned automatically when you create a site | | Custom domain | Optional branded URL bound to a site or workspace | | Community learner | External role for people who sign up through a site: limited permissions, site-scoped access, excluded from workspace seat counts | | External group | Per-site group community learners are enrolled into automatically | | Publish / unpublish | Controls public accessibility; unpublishing revokes external access | ## Typical deployments | Organization | How Sites is used | |---|---| | Corporate L&D | Public onboarding previews for candidates; a partner certification portal on their own domain | | Trade unions and associations | Member continuing-education catalog; apprenticeship resources open to prospects | | Higher and continuing education | Alumni learning library; non-credit storefront separate from the formal LMS | | Training vendors and consultancies | White-labeled delivery for clients without granting workspace admin access | | Nonprofits and community organizations | Free public courses, with gated advanced tracks behind signup | | Developer relations | A public API and SDK academy: docs that teach, with hands-on activities and an in-context tutor | | Enterprise SaaS | A customer education hub on your own domain rather than a third-party platform | ## What Sites is not - **Not a separate product.** It is a layer on the workspace you already have, not a second content silo. - **Not anonymous progress tracking.** Public browsing is read-only until login. - **Not a full website builder.** You configure navigation, footer, branding and course layout, not arbitrary CMS pages. Your marketing site stays your marketing site. - **Not a replacement for internal cohort management.** Groups, assignments, grading and analytics for enrolled classes stay in the workspace. ## Standalone activity links Separately from Sites, any topic or single activity can be delivered on its own: no course navigation, just the work and a single onward action when it is complete. Progress still counts toward the parent course. This is what makes a Honen activity embeddable inside a chat message, an email, an onboarding flow, or another product. --- # LMS Integrations Honen is a full learning management system and can replace one outright, but it does not have to. It also delivers into an LMS you already own, which means adopting it does not require a migration. There are four shipped paths, and they are independent and simultaneous. ## Choosing a path You do not need to know which standard your LMS speaks. Pick the platform by name. | You run | Path | Why | |---|---|---| | Canvas, Moodle, Blackboard, D2L Brightspace, Skilljar | LTI 1.3 | Deepest integration: automatic rosters, server-side grades | | Cornerstone, Docebo, SAP SuccessFactors, Workday Learning, TalentLMS | SCORM | The package-based corporate systems | | Google Classroom | Classroom add-on | Native, and better than LTI there | | A Learning Record Store (Watershed, Learning Locker, SCORM Cloud) | xAPI or cmi5 | You want the full learning record, not one score per course | | Anything else, or not sure | SCORM | Near-universal. Not being listed does not mean it will not work | ## What each path guarantees | | LTI 1.3 | SCORM | Google Classroom | |---|---|---|---| | Setup | Admin handshake per platform | Download a package, upload it | Install the add-on | | Rostering | Automatic | Learners created on first launch | From Classroom | | Grade passback | Server to server | Browser-side, through the LMS's SCORM API | Server to server, as a draft grade | | Instructor picks content inside the LMS | Yes | No | Yes | | Re-deploy after a content edit | Never | Never | Never | ## LTI 1.3 and LTI Advantage Connecting is, in practical terms, pasting a single URL. Once connected: rosters are read from the LMS so there is nothing to import; groups and cohorts are created from the LMS structure; learners never sign in to Honen separately because they are already authenticated; and instructor, TA, learner and admin roles map into Honen roles. Courses can run fully embedded, so the entire experience including the AI tutor runs inside the LMS without anyone leaving, or as a one-click launch into a new tab already signed in. Grades flow into the native gradebook, including AI-graded mastery rather than a completion checkbox. Deep linking lets instructors pick content from inside the LMS, and dynamic registration is supported. ## SCORM 1.2 and 2004: thin, live packages You get the portability of SCORM without the frozen-snapshot tax that normally comes with it. A Honen SCORM package does not contain a copy of your course. It carries a small launcher that loads the live course inside the LMS. Edit the course in Honen and every deployed package reflects the change immediately. You upload once and never re-upload for a content change; only a settings change (audience, group, reporting policy) requires a fresh export. ## Google Classroom A native add-on rather than an LTI deployment, with attachment discovery inside Classroom, graded coursework, ungraded links, materials and announcements. Grades are written as a draft for the teacher to publish, scaled to the Classroom assignment's point value. Copying a Classroom course carries the assignment scope and criteria across, so two sections are never graded against different work. ## xAPI and cmi5 Three distinct capabilities: - **Honen produces xAPI.** xAPI 1.0.3 statements are generated server-side for course, topic, activity, test and project events, stored in Honen, and optionally forwarded to your Learning Record Store with retry. - **Honen serves an LRS.** Honen exposes the xAPI HTTP API (statements, activity state, activity and agent profile, activities, agents, about), authenticated with workspace-scoped credentials an administrator mints. - **cmi5 export.** Thin packages that iframe the live course, provision the LMS learner, and relay the cmi5 statement sequence to the LMS's LRS while Honen keeps the same statements. Both require an organization workspace and are available by default: the first cmi5 export materializes the configuration and provisions the launch tenant, in the same way the first SCORM download does. ## Running several at once Nothing is exclusive: - One workspace holds as many LTI connections as it needs, each with its own keys, role mappings and grade policy. - The same course can ship as SCORM 1.2 to one client and SCORM 2004 to another, plus packages targeting internal members and external learners separately. Each carries its own signed secret, so revoking one client's access does not affect the others. - Separate tenants keep clients isolated and can bind a cohort to its own group. - Exporting a package requires no LMS connection at all. A single workspace can be LTI-connected to two universities, have SCORM packages in five corporate systems, and run Google Classroom, all serving the same live courses into the same analytics. ## Honen stays the source of truth Every integration receives a projection of learner progress. No path lets an external system write learning state back into Honen. The LMS gets a compliance-friendly gradebook entry; Honen keeps the full mastery picture, with SCORM, LTI and xAPI learners appearing in analytics alongside native members. ## Seats Learners provisioned through any path are either internal members, who are normal workspace members consuming a seat, or external learners, who are attached to a Honen Site with a capped role and excluded from seat counts. Either way they are created automatically on first launch and can claim a full account later. ## Boundaries - SCORM gradebook writes happen from the browser while the learner has the course open. LTI and Classroom are server to server. - Progress percent as a score policy is coarse: not started, started, complete. - Certificates are issued per curriculum rather than per course. - Session isolation is per tenant, not per learner, so two learners launching the same package in the same browser profile on a shared device share a session. Route cohorts to separate tenants, or use separate browser profiles. This matters for shared-device deployments. - Package export requires an organization workspace. - LTI 1.1, Common Cartridge and AICC are not supported. There is no importer, exporter or runtime for any of them; LTI 1.3 is the supported version. - Honen produces xAPI and serves an LRS but is not an xAPI content player. Importing or playing a third-party xAPI or cmi5 package inside a Honen course is not supported. - The LRS authenticates with HTTP Basic. OAuth 1.0a and signed statements as an authentication path are not implemented. - A course exports as a single cmi5 assignable unit; per-topic units are not produced. - Learners arriving without an LMS identifier are still tracked separately but appear as generated identifiers rather than names. --- # Analytics and Reporting Honen does not just track completion. It reports how an organization is actually learning, and it lets you reach any answer by asking in plain language. ## Ask, and the AI builds the report Describe the report you want and the analytics agent does the rest. There is nothing to configure and no query language to learn. It analyzes the data across your organization, decides how best to present it, and produces a report page with charts that lives on your dashboard and refreshes itself. - **Ask in plain language.** "Show me which units the East region is struggling with this quarter." Type the question, get the report. - **The AI builds the charts.** It picks the visualization, builds the charts and lays out a clean report page automatically. - **Each answer becomes a permanent page.** Pinnable, revisitable and shareable, and it keeps itself current, rather than being a one-off export. - **It analyzes rather than just displaying.** It surfaces what matters and says where learners are doing well and where they are stuck. Refresh cadence matches the decision: minutes for a live cohort, hourly for aggregate KPIs, daily for a board report, monthly for a funder submission. Each refresh runs a deterministic script that can be reviewed, which is what makes the output usable for an audit rather than only for internal curiosity. Representative requests: - "Certifications expiring in the next 90 days, grouped by manager." - "Time-to-complete on the new-hire track by start month." - "Compliance pass rate by region, with a manager drill-down." - "Managers whose team has zero completions on the Q4 rollout." - "DFW rates by section." / "At-risk students four weeks before state testing." - "Journeyman-exam pass rate by local by year." - "Pre-apprenticeship to apprenticeship conversion rate by area." ## The depth underneath Real comprehension, not seat-time checkboxes: - **Per-learner mastery**: what each person actually understands, not whether they finished. - **Test and assessment scores** across people, cohorts and topics, down to individual questions. - **Time spent on each activity**, so you can see where attention goes and where it drops. - **Project and hands-on performance**, including submissions and pending reviews. - **Which units and topics learners struggle with overall**, so you fix the content instead of chasing the learner. - **What is landing well**: the formats and modules that work, so you can do more of them. - **Cohort and program views** across teams, locations, classes, locals and agencies, with group rollups and needs-attention flags. Course analytics use the assigned cohort as the denominator rather than everyone who happened to open the course, and recompute completion live against current criteria, so tightening a bar later does not leave false completions behind. Scoped to an assignment, figures are measured against that assignment's bar. Because activities complete on real signals rather than a "next" click, completion means the material was genuinely engaged with. Every quiz and test answer also carries a confidence check, which distinguishes a learner who knows the material from one who guessed and one who is confidently wrong. See [the learner experience](https://honen.com/llms/learner-experience.md). ## Ready for whoever asks Because reports are built on demand from a question, you can produce exactly what a given audience needs in seconds: a leadership standup view, a board review, or outcome reports formatted for regulators and funders. Same data, any framing, no analyst queue. For the statutory formats a publicly funded program has to file, see [funder and compliance reporting](https://honen.com/llms/funder-and-compliance-reporting.md). ## Operations in the same conversation The workspace assistant that answers analytics questions also runs the workspace: inviting people, creating groups, and managing assignments. Destructive steps require confirmation, and reads that expose roster-wide data require management permission. See [the three AI surfaces](https://honen.com/llms/ai-surfaces.md). ## Terminology and branding Every workspace renames what its users see, and the renaming applies across every analytics surface, so reports speak the organization's own language. See [people, groups and workspaces](https://honen.com/llms/people-and-workspaces.md). --- # Funder and Compliance Reporting Publicly funded training carries a reporting burden that has little to do with teaching. Historically it means analysts hand-assembling spreadsheets from an LMS export, an HR system and a state wage file: slow, error-prone, and liable to fail at audit. Honen compiles the reports a program has to file straight from the delivery data it already holds. You pick a report, and it gets built. ## The honest positioning **Honen auto-fills every field it owns, shows exactly which fields still need an intake upload or a wage match, and produces the funder's file.** It is not "Honen files your WIOA report". The reporting obligation is usually the customer's: for federal programs the grantee, state agency or registered provider is the legal submitter. Honen is the delivery and evidence layer that feeds them clean data. Two data gaps recur across almost every program, and no learning platform can close them: post-exit employment and earnings, which come from a state unemployment-insurance wage match, and intake demographics and eligibility. Honen ingests these rather than pretending to generate them, and flags them clearly so nothing ships blank. What Honen genuinely powers is **measurable skill gain** (through progression, milestones, exams or credentials) and **credential and certificate attainment**. Both are native. ## Four layers 1. **Own-data layer.** Everything Honen already knows: enrollment, contact and training hours, completion, measurable skill gain, credentials, and the exact curriculum version delivered, compiled into a canonical per-participant record keyed to a stable participant ID, never a Social Security number. 2. **Import layer.** The data Honen cannot know: demographics, eligibility, employment, wages and on-the-job-training hours, uploaded by CSV and matched to the right learner by email. SSN and TIN columns are detected and never imported. 3. **Mapper layer.** The merged record is transformed into each target format's file: fixed-column CSV for PIRL, aggregate tables for NRS, evidence packs for ETP. 4. **Readiness and gap flagging.** Before anything generates, a readiness check shows precisely which required fields are missing and for how many participants, with per-field counts such as "missing for 1 of 10". You fix one record instead of re-running the whole report, and nothing is ever silently emitted incomplete. ## Formats supported | Format | Program | Notes | |---|---|---| | **PIRL** | WIOA Title I, to WIPS | Fixed-column per-participant record, leading zeros preserved, no SSN | | **NRS** | Adult education (Title II) | Aggregate tables | | **RAPIDS** | Registered apprenticeship | Related technical instruction and competency data to a PIRL-format extract | | **Perkins V** | CTE Consolidated Annual Report | Concentrator completion, credential attainment | | **ETP** | California Employment Training Panel | Trainee evidence pack: hours, completion, rosters, curriculum version | | **ETPL** | Eligible Training Provider List | Program and cohort-level outcomes | Accreditation and institutional-research exports are covered for education deployments. ## How it helps - **Auto-filled from delivery.** Enrollment, hours, completion, skill gain and credentials come straight from what learners actually did. No manual tallying. - **Curriculum version stamped.** Every completion records the exact curriculum revision delivered, so an auditor can see what was taught, when, and to whom. - **Easy imports.** Demographics, wages and eligibility upload by CSV and match by email. Blank templates and filled examples round-trip with no manual column mapping. - **Guided, not a wall of errors.** The readiness check walks through what is ready versus what needs importing. Import only the missing records, or generate with blanks intentionally. - **Scoped reports.** Pull a report for one group, one employer, or the whole workspace from the same data. - **Audit-clean output.** Each format emits exactly the columns its system expects, aggregate formats roll up correctly, and an empty scope never produces a misleading zero file. - **Available conversationally.** The workspace assistant can list formats, run readiness checks, set learner records and compile reports, gated on reporting permission. ## Who it is for - **Government agencies** running WIOA, adult education, apprenticeship or CTE programs file the heaviest stack: PIRL and WIPS, NRS, RAPIDS, Perkins, plus state ETP and ETPL. Enrollment, hours, completion, skill gain and credentials fill themselves, and reports scope to one cohort, an employer group, or the whole agency. - **Higher education and community colleges** report Perkins V concentrator outcomes and maintain ETPL eligibility. Completion, credential attainment and program performance come from real coursework, and the same records document value-added earnings for state boards, including the case for short-term programs. - **Enterprise L&D** drawing on incumbent-worker grants, apprenticeship funding or state training dollars needs reimbursement evidence: training hours per employee, completion, rosters and the curriculum version delivered. For registered apprenticeships, related technical instruction completions and competency milestones flow into the RAPIDS and PIRL extract. - **Any grant-funded training provider.** If funding depends on documented outcomes, Honen turns delivery into the funder's file and flags the intake and wage data you still need to supply. ## Security and privacy Access is gated by a reporting permission held implicitly by workspace administrators, and every sensitive action is audited. SSN and TIN are scrubbed on the way in, by column name and by value shape, and never persisted. All data is workspace-scoped: generation resolves participants against real membership, and downloads re-check ownership on every request. Report files download through an authenticated route with no long-lived public URLs, and expire on a retention schedule. Reporting is deliberately kept outside connector and messaging-bot capabilities, so a scoped bot can never reach this data. ## Cadence and custom rollups Cadence is set per report rather than globally: monthly for a grant cycle, weekly for cohort health, daily during a placement push, hourly during a testing window. Each stakeholder gets the view they need at the interval they need it. Anything not covered by a standard format can be described in a sentence and pinned as a permanent auto-refreshing report: outcomes by cohort and area, time-to-certification by trade, completion ratio for a target population by quarter, conversion from pre-apprenticeship to apprenticeship. See [the three AI surfaces](https://honen.com/llms/ai-surfaces.md). ## Good to know US funder schemas are reissued periodically, including PIRL annually, the NRS tables and the Perkins reporting guide. Confirm the current version from the source before building a given submission. --- # People, Groups and Workspaces ## Workspace types A **personal workspace** is for one person learning or teaching on their own. An **organization workspace** adds groups, roles, assignments, workspace analytics, package export for LMS delivery, and Agents. Organization workspaces are isolated from each other at the data layer. A deployment serving many clients can additionally use separate tenants so that each client's learners, sessions and credentials stay independent, and revoking one client's access does not touch another's. ## Getting people in - **Invite by email**, individually or in bulk, including from a pasted list or a CSV. - **Bulk roster import** for organizations, covering instructors, classes and learners. - **Single sign-on** over SAML 2.0 and OIDC, so access follows your existing directory. - **Clever and ClassLink** for K-12 district rostering. - **LMS launch.** Learners arriving over LTI 1.3 are rostered automatically from the LMS, and learners arriving through a SCORM or cmi5 package are created on first launch. Nobody has to be imported by hand. - **Self sign-up** through a public [Site](https://honen.com/llms/sites-and-distribution.md), for community and academy audiences. Learners created automatically through an LMS or a Site can claim a full account later without losing progress. ## Roles Roles run from member through instructor to administrator, and permissions can be scoped to a subtree of groups. A group-scoped instructor reaches the groups they manage and their descendants, and nothing else. Individual capabilities are separately gated, including whether someone can manage assignments, read roster-wide analytics, create courses, or administer the knowledge base. Instructors hold assignment management by default; members do not. ## Groups Groups nest arbitrarily, and the nesting is what makes large deployments manageable: a district contains schools which contain classroom periods; a union contains locals which contain apprenticeship classes by year; a company contains regions which contain teams. Targeting anything at a group always includes its descendants, so an assignment to a school reaches every period inside it. Groups can be created conversationally, and a new member joining a group inherits every assignment already targeting it. ## Membership status A member can be marked **inactive** without being deleted and without losing progress. This exists because real programs need it: a laid-off worker, a student between terms, an employee on leave. When they return, their history is intact. Cohorts can be **retired** cleanly at the end of a term, year or program, without resetting anyone's earned progress or disturbing the audit trail. ## Curricula and certificates A **curriculum** is an ordered set of courses that together represent a qualification: an apprenticeship year, a certification track, a degree pathway, an onboarding program. Certificates are issued at the curriculum level rather than per course. Finishing an individual course, or a scoped assignment within one, earns credit, analytics and an LMS grade, but the certificate belongs to completing the curriculum. Certificates are verifiable, carry unique IDs, are shareable to LinkedIn and downloadable as PDF, and are backed by AI-graded mastery across quizzes, projects and simulations rather than time spent. ## Terminology Every workspace can rename what its users see. "Group" can become cohort, section, local, department or crew; "member" can become student, employee, apprentice or participant; "instructor" can become teacher, trainer or journeyworker. The renaming is applied consistently across the product, including every assignment and analytics surface, so the platform speaks the organization's own language rather than making people learn a new one. ## Branding Three layers, used together or separately: 1. **Brand defaults** seed the visual identity of AI-generated course imagery: name, logo, colours and free-text guidelines. 2. **White-labeling** replaces dashboard chrome and the login experience on your own domain. 3. **Per-site branding** for each public storefront, so different audiences can see different presentations of the same catalog. ## Notifications Delivery reaches people where they already are: in-app and email, plus Slack, Microsoft Teams, Google Chat and iMessage. Administrators control which channels push, and individuals can opt out. Automatic due-soon and overdue alerts fire on thresholds the administrator sets, once per person. Streak milestones are emailed at 3, 7, 14 and 30 consecutive days. --- # Knowledge Base and Parallel Training A workspace knowledge base holds the approved material an organization teaches from: policies, standard operating procedures, manuals, product documentation, codebooks. It serves two audiences from one corpus. ## Parallel training The same approved material trains the humans and answers the machines: - **People** learn it as structured courses, with progress, assessment and certificates. - **Agents** answer from it on Slack, Microsoft Teams, iMessage and your own website. See [Agents](https://honen.com/llms/agents.md). - **Your own AI tools** reach it over MCP, so Claude, Cursor and internal bots work from the same current material. One permission model, one update path. When a policy changes, the course and the assistant both reflect it, rather than drifting apart. ## What it holds Documents and folders with tags, versioning and full revision history, plus images that are searchable alongside text. Documents can be shared to the whole workspace, to groups, to named users, or through public links. Roles run from viewer through editor and maintainer to admin. Screen recordings can be imported and turned into structured visual walkthroughs. ## Feeding courses Knowledge base documents can be pulled directly into course creation, where they are snapshotted into the course. When the original document later changes, the course raises an alert with one-click refresh, so a course never silently drifts from the policy it was built on. Readings grounded in knowledge base sources carry inline citations back to the document, with visibility configured per course. See [the learner experience](https://honen.com/llms/learner-experience.md). ## Integrations and MCP Content comes in from files, URLs, YouTube, public GitHub repositories, Granola meeting notes, and connected MCP servers including Linear, Notion, HubSpot and Google Drive. Honen also exposes its own MCP interface, so an organization's existing AI tooling can read and write the knowledge base, read and edit courses, and drive workspace operations under the same permissions a person would have. --- # Agents An Agent is a custom AI assistant that belongs to a workspace. It has one prompt, one set of guardrails, one access policy, and one or more channels where people can talk to it. Build it once, then deploy the same behaviour to Slack, Microsoft Teams, iMessage, and an embedded chat on your own website. This is what turns a knowledge base from something people study into something that answers them at their desk. ## Anatomy | Part | What it controls | |---|---| | Prompt | The Agent's role, goals, tone and response style | | Guardrails | Non-negotiable limits, applied after the prompt and overriding it | | Connector | The maximum data and actions available to the Agent | | Identity mode | Whether it acts as a service, as the person asking, or both | | Channels | Slack, Microsoft Teams, iMessage, website embed | | Status | Draft, active, disabled or archived | ## One Agent, every channel Channels are deployments of one Agent, not separate Agents to keep in sync. Adding Slack and Teams to the same Agent gives both the same prompt, guardrails, permissions, approved knowledge and courses, and lifecycle. Credentials and routing stay isolated per channel; behaviour does not diverge. Changing the policy once applies it to every active channel. - **Slack.** Mention the Agent in a channel or send it a direct message. Replies stay in the thread. - **Microsoft Teams.** Private chats and standard channel threads. Mention it once to start a thread and it keeps answering there. - **iMessage.** Text a question and get an answer grounded in the current material. One active iMessage Agent per workspace. - **Website embed.** Drop it into your own site as an iframe, public or authenticated. ## Permissions are separate from instructions **An Agent cannot exceed its Connector's permissions, even if its prompt or a user asks it to.** Prompts describe intended behaviour; the Connector decides what is technically possible, and it is enforced server-side. Guardrails are stored separately from the prompt and take precedence when the two conflict, so a limit like "never reveal private employee data" is not one instruction competing with a long persona description. Three quick profiles cover most cases: | Profile | Access | |---|---| | Answer questions | Read approved knowledge and courses | | Build courses | Read knowledge, create courses, edit course content | | Run learning operations | Read people and analytics, manage assignments | Beyond these, access is granular per capability: knowledge base (read, write or admin), courses, course creation with daily and lifetime caps, people, analytics, assignments, groups, monitors and alerts, workspace curation, workspace settings, and the integration audit log. Each can be narrowed by an allowlist of specific files, folders, tags, per-path rules or named courses. Where path rules overlap the most specific wins, so one file can stay read-only inside a writable folder. The recommended approach is the smallest policy that supports the Agent's job, adding capabilities only when a real workflow needs them. ## Identity - **Service.** The same access for every sender, no workspace membership required. For public support and anonymous audiences. - **Hybrid.** Links a sender to a workspace member where possible and falls back to the service identity otherwise. - **Member gated.** The sender must resolve to a current workspace member. This is the mode for internal assistants touching private information, and membership is re-checked on every session. The Connector remains the ceiling in every identity mode. Identity narrows access; it never widens it. ## Embedding on your own site A public Agent takes three steps: enable public access, list the website origins allowed to embed it, and copy the generated iframe. For signed-in users, your server mints a short-lived Agent URL for the current person, so the assistant knows who it is helping without you managing another login. Those URLs carry a single-use ticket that expires in two minutes and is never exposed in URLs, logs or referrers. The server key stays on your server. ## Lifecycle Agents move through draft, active, disabled and archived. Disabled is a clean pause that keeps the configuration. Archiving is permanent and thorough: it disables every deployment, revokes embed keys, tickets and live sessions, releases channel routing, and revokes the managed access policy. Conversation history is isolated by Agent, channel, conversation and thread. Permissions are enforced server-side, guardrails override conflicting instructions, provider credentials are sealed at rest and never returned to the browser, every inbound Slack and Teams request is cryptographically verified, embed origins are checked when a session starts and again while it runs, and member-gated sessions re-check current membership. ## Suggested starting policies | Agent | Policy | |---|---| | Internal knowledge assistant | Answer questions, member gated | | Public support Agent | Answer questions, service, approved public knowledge only | | Course creation Agent | Build courses, member gated | | Operations Agent | Run learning operations, member gated | ## Boundaries - Slack and Teams channels remain in draft until provider setup is complete. Selecting a channel is not the same as activating it. - A custom Teams Agent requires an Entra application, an Azure Bot registration and a Teams app package, so it involves IT. Slack is considerably lighter. - Only one Agent can be the active iMessage responder per workspace at a time. - A workspace-wide Slack or Teams installation is not automatically attached to every custom Agent. The default workspace assistant and a custom Agent are separate things. - Agent-managed access policies cannot be shared between Agents; reusing one means attaching an existing Connector explicitly. --- # Integrations and MCP Honen connects to the tools where your knowledge already lives, keeps it current on its own, and lets AI assistants work with it, yours and your own. Almost every question about integrations becomes easy once you know there are three separate things, pointing in different directions. | | What it means | Direction | |---|---|---| | **1. Bringing content in** | Point Honen at documents you already have and it pulls them into the knowledge base and keeps them updated | Your tools into Honen | | **2. Honen's AI reaching into your tools** | The assistant is given access to your working systems, so it can pull a ticket or check a calendar while it helps you | Honen into your tools | | **3. Outside AI reaching into Honen** | Claude, Cursor or another assistant connects to Honen so it can search and edit your knowledge base from where your people already work | Other AI into Honen | Most customers mean the first when they say "integrations". The second and third are the AI-assistant story. ## 1. Bringing content in | Source | What it pulls | |---|---| | Google Drive | Docs, PDFs, slides, spreadsheets, images, and whole folders including sub-folders | | Notion | Pages and their sub-pages | | Granola | Meeting notes: the AI summary, the full transcript, or both | | Any website | Crawls a site or a section and turns the pages into knowledge | | YouTube | Paste a link, get the transcript as a document | Files can also be uploaded directly (any type, up to 1 GB each), along with browser screen recordings, which become interactive walkthroughs rather than videos, and public GitHub repositories. **Setup effort differs by source, and the difference is real.** Google Drive and Notion are a single sign-in with your normal account, then browse and pick. Granola takes one key from your account, checked on the spot so it fails immediately rather than silently later. Websites and YouTube need no connection at all: paste a link. These are available by default in every workspace and an administrator can turn them off. For folder-based sources you can link a single document or an entire folder, and everything inside comes along. ### Staying current without anyone remembering Honen re-checks connected sources about every minute. When something changed upstream it updates the copy; when nothing changed it does nothing, so watching a large folder costs nothing. Edit a Google Doc and the course material built on it reflects the edit without anyone filing a ticket. Deletions propagate too: remove a file from the source and it comes out of the knowledge base. This is a fast poll rather than a live stream, which is worth being precise about. ### Two details that matter later **Images do not rot.** Many tools serve images through links that expire within hours. Honen copies every image into your workspace's own storage at import time, so a document that looks right today still looks right in a year instead of filling with broken images. **Scanned documents become searchable.** A PDF that is really photographs of pages goes through the same text-recognition pipeline as a hand-uploaded file, so it is searchable and usable as course material. ## 2. Honen's AI reaching into your tools Once connected, the assistant can do things that span tools ("summarize this week's Linear tickets for the onboarding course") instead of people copying context back and forth. Available today, one click each: | Category | Tools | |---|---| | Productivity | Linear, Notion, Granola | | Sales | HubSpot | | Communication | Slack | | Google | Calendar, Chat, Drive, Gmail, Contacts | Teams can also connect their own internal tools, not only the ones listed, provided those tools speak MCP. **If one of these tools is slow or down, the assistant keeps working.** It carries on without that tool for the moment rather than hanging on someone else's server. ## 3. Outside AI reaching into Honen Honen works as a connectable source for assistants like Claude and Cursor, so someone who lives in one of those tools all day can search the knowledge base, read a document or update one without switching to Honen. Course reading and editing and workspace operations are reachable the same way. Two ways to set it up, and the distinction matters: - **Sign-in connection.** A person connects with their own Honen account and gets exactly the access they already have. Nothing new to manage. - **Scoped connection.** An administrator creates an access token limited to a specific slice: one folder, one course, read-only, assistant-only. This is the route for contractors, agents and automations that should see one thing and nothing else. ### About MCP MCP is the open industry standard for how AI assistants connect to outside tools. Think of it as the USB-C of AI integrations: one shape, works everywhere, no custom adapter per tool. Honen speaks it in both directions, which is unusual: Honen's AI can plug into your tools, and your AI can plug into Honen. Being on a standard is also why the list of available tools grows without per-vendor engineering. This is what makes [parallel training](https://honen.com/llms/knowledge-base.md) possible: one approved corpus trains the humans as courses and grounds the machines as skills, with one permission model and one update path. ## 4. Honen's assistants into your channels A workspace can also build its own [Agents](https://honen.com/llms/agents.md) and deploy them to Slack, Microsoft Teams, iMessage and an embedded chat on its own website. One prompt, one set of guardrails, one access policy, every channel. ## Control and safety - **Access is scoped, not all-or-nothing.** A connection can be limited to a single folder, and can be read-only even inside a folder whose parent grants write access. Anything not explicitly granted is denied. - **Nothing is shared by accident.** Content pulled in follows your existing knowledge base permissions. Connecting a source does not publish it. - **Everything is logged.** Every connection, every token issued and every action taken through one is recorded with who, what and when. - **Revoke instantly.** Any connection or token can be shut off and access stops immediately. - **Refreshes run as the person who linked the source**, so a synced folder can never show more than that person could see themselves. - **Automated access expires by default.** Credentials created for short automated tasks last minutes, and can never be more powerful than the person the AI is acting for. ## Identity and delivery - **Single sign-on** over SAML 2.0 and OIDC. - **LTI 1.3** for LMS-authenticated launches, with automatic rostering and grade passback. - **SCORM 1.2 and 2004**, **Google Classroom**, and **xAPI and cmi5** into a Learning Record Store. See [LMS integrations](https://honen.com/llms/lms-integrations.md). - **Clever and ClassLink** for K-12 district rostering and sign-in. - **API access** for issuing assignments from events in another system, so a new hire in an HR system or a role change can trigger the right training automatically. ## Common questions **Does connecting a source move my files out of the original tool?** No. The original stays where it is and remains the source of truth. Honen keeps a current copy. **If I edit the copy in Honen, does it push back?** No. Sync flows one way, inward. Nothing you do in Honen can damage the original. **How fast do changes show up?** Typically within a minute or two of the upstream change. **Can one person's connection expose their private files to the whole team?** Only if they choose to share what they linked. Linked content follows normal knowledge base permissions. **Can we connect our own internal tools?** Yes, if they support MCP. No custom build is needed on Honen's side. **What happens if a connected tool goes down?** Content already in Honen is unaffected, because it is a copy rather than a live window. Syncing resumes on its own. --- # Security and Compliance Compliance and privacy here are architectural rather than procedural. Data isolation, audit trails, permission ceilings and assessment integrity are properties of how the system is built, not policies applied on top of it, which is why they hold even when an AI agent or an end user asks them not to. ## Certifications and frameworks - SOC 2 Type II - FERPA compliant - COPPA compliant, for K-12 use - FedRAMP-ready posture for government deployments - AWS GovCloud (US) deployment available for government ## Data handling - Complete per-workspace data isolation - AES-256 encryption at rest - Full audit logging - Enterprise data is not used to train AI models - Honen does not sell or share learner data ## Identity - SAML 2.0 and OIDC single sign-on - LTI 1.3 for LMS-authenticated launches - Signed short-lived URLs for embedding an assistant for an already-signed-in user - Roles and group-scoped permissions, so an instructor who manages some groups reaches only those groups and their descendants ## Agent and assistant safeguards Permissions for AI Agents are enforced server-side and independently of the prompt. An Agent cannot exceed its access policy even if its prompt or a user asks it to, and guardrails are stored separately from the prompt and take precedence when the two conflict. Provider credentials are sealed at rest and never returned to the browser, inbound requests from Slack and Microsoft Teams are cryptographically verified, embed origins are checked when a session starts and again while it runs, and member-gated sessions re-check current membership. Archiving an Agent disables every deployment, revokes embed keys, tickets and live sessions, releases channel routing, and revokes its access policy. See [Agents](https://honen.com/llms/agents.md). ## Assessment integrity During a test, the AI tutor's search and image tools are withdrawn, enforced both in the agent and on every request. Attachment reading is disabled whenever a learner has an unsubmitted test anywhere in the course. ## Content provenance Readings grounded in knowledge base sources carry inline citations back to the source document, with six levels of per-course visibility. Authors are warned when a cited passage no longer matches the current source. Course changes are revisions with full history and rollback, and knowledge base documents are versioned.