Current limits

Check these limits when planning a course or integration. They describe this authoring build, including the distinction between reviewing a course and delivering it to learners.

Delivery

Publish gives a course a public web link with practice-only quizzes and no learner tracking (see Publish a course to the web). SCORM and xAPI exports and public enrollment aren’t available. With an administrator-configured Canvas LTI connection, learners can open published courses in Canvas and scored quizzes can send grades back. Sharing for review provides reviewer access, not learner delivery.

Native course packages are available for full-course export and import. The current UI transport is limited to 32 MiB. Packages include a player for a local HTTP server or static hosting; standalone quizzes are practice only, without trusted grading or LMS reporting. Selective import and arbitrary HTML/SCORM import are not available.

Structure

Courses store structured content. Columns can contain blocks, and accordion items can contain one level of tabs. Phone and tablet layouts override the desktop layout while sharing the same content.

AI

AI Assist works on one page at a time — the selected text, selected block, or the current page — and never edits the cover. The server turns your instruction into validated ops; nothing changes until you choose Apply, then the change saves as a labeled AI entry in version history and undoes in one step. Proposed changes can’t be applied if the course changed while the assistant worked; your edits are kept.

Blueprint building turns an AI course plan into ops that keep the blueprint’s objectives and links, building only pages that still need creating. A blueprint page can be a cover, content page, or quiz; the blueprint kind takes precedence over the plan.

Cancellation depends on the workflow. AI Assist can stop accepting a late result after cancellation, but that doesn’t establish that the provider request stopped or was refunded.

Validation and evidence

Review and Alignment help identify specific issues. You still need to check content accuracy, accessibility, and learning design. Use current feature behavior when an older acceptance report describes something different.

Access

Private media and uploaded fonts use signed links that expire. Reviewers don’t gain authoring rights. Agent integrations act with the signed-in caller’s permissions.

File upload limits

During beta, a file may be up to 200 MB. ZIP unpacking allows up to 500 files and 1 GB total uncompressed content; each extracted file must also fit the per-file limit. Keep a ZIP as one downloadable archive when appropriate. Replacing or restoring a shared file updates every course using that resource.

File preview limits

Image and PDF previews are limited to 50 MB per file. DOCX previews are limited to 32 MB per file. Files over these size limits can still be downloaded. Unsupported formats and oversized files show a message instead of a preview.

Integration retry limits

Course operations and blank workspace creation support idempotency keys. AI course drafting does not currently provide the same replay guarantee, even when the HTTP create request includes a key. After an interrupted draft request, inspect the course list before retrying. See HTTP integration.

Validation context

MCP validate_course uses the shared validator but currently does not receive the editor’s complete resource set. Run the editor’s Review panel as well when checking missing private files; do not treat an MCP result as proof of resource availability.

Managed HTML components

The inline-v1 profile supports self-contained HTML/CSS/classic scripts and managed assets. Limits are 120,000 source characters, 32 assets, 5 MiB per asset, and 10 MiB total. It does not support ZIP/module loading, external services, persistent learner state, or arbitrary HTML/ZIP conversion. Supported managed components can be included in a native course package. Asset bytes are verified before rendering; replacing a bound file requires attaching the new version and updating the source token. Component controls and limits.

Cancel an in-progress AI request

While AI Assist shows Brewing suggestions…, choose Cancel to stop accepting the result. The panel confirms Request canceled. Your content is unchanged. Canceling does not establish that the provider request stopped or was refunded.