Tools

List every course with its id and completion progress. Call this first to discover course ids for scoping.

Hybrid semantic + keyword search across lesson content (local embeddings, no API key). Returns the most relevant sections.

List lesson titles within a course, optionally filtered to one module.

Fetch the full markdown body of one lesson by lesson_id.

Fetch one lesson section by lesson_id + anchor, as returned by search_lessons. Use to read a hit in full when the preview was truncated.

Create a new, empty course and return its course_id. Call this first when building a course from scratch, then add_lesson / add_lessons_from_markdown.

Append ONE markdown document to a course as a single lesson. If title is omitted, the document's H1 is used — same parsing as the GitHub crawl, minus the fetch.

Split ONE markdown document into MULTIPLE lessons (H2 → module, H3 → lesson, the single-README crawl rule) and append them all to a course.

Walk a local NN-module/NN-lesson/README.md tree on the server's filesystem and import every README as a lesson — the GitHub crawl, minus the fetch. Omit course_id to create a new course (requires title); pass course_id to append (never destructive).

Create a flashcard (front question + back answer) on a lesson. Appears in the lesson's deck immediately.

Append a timestamped markdown note entry to a lesson's notes.

Capture a YouTube video as a video lesson: fetch its timestamped transcript (captions), store the raw segments, and write a transcript scaffold as the body. Re-capturing the same video reuses the existing lesson.

Fetch a video lesson's raw timestamped transcript (one [h:mm:ss] line per segment) — the source for authoring.

Overwrite a lesson's markdown body (the authoring write-back: replace a video lesson's transcript scaffold with a synthesized, research-enriched lesson). Place an HTML comment `<!--t:SECONDS-->` after a heading to make that section seek the player.