> ## Documentation Index
> Fetch the complete documentation index at: https://nirmauniversity-d6cdc5d7-dev.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Pipeline and output

> Understand how NoteWise processes YouTube inputs and writes output artifacts.

The processing path is implemented by `CorePipeline` and the pipeline execution helpers.

## Processing flow

<Steps>
  <Step title="Cache check">
    Unless `--force` is set, NoteWise checks the SQLite cache before metadata or
    transcript work.
  </Step>

  <Step title="Metadata and transcript">
    NoteWise resolves YouTube metadata, captions, transcript segments, chapters,
    and playlist context when applicable.
  </Step>

  <Step title="Generation">
    Transcript text is chunked with source defaults `4000` size and `200`
    overlap, then sent to the configured LiteLLM model.
  </Step>

  <Step title="Artifacts">
    NoteWise writes notes, optional quizzes, optional transcript exports, and
    rendered formats.
  </Step>

  <Step title="Persistence">
    Run stats, output paths, transcript content, tokens, costs, and timings are
    persisted to SQLite when possible.
  </Step>
</Steps>

## Chapter behavior

Videos with YouTube chapters use chapter-aware generation. Without `--chapter-directory-output`, the result is still bundled into a final notes file. With `--chapter-directory-output`, NoteWise writes per-chapter Markdown files into a folder and stores `.notewise-output.json` ownership metadata.

## Output formats

| Format | Behavior                                                                                                                                                                                                                         |
| ------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `md`   | Raw Markdown notes. Default.                                                                                                                                                                                                     |
| `html` | Markdown rendered with built-in CSS.                                                                                                                                                                                             |
| `pdf`  | Uses `fpdf2` with a bundled Noto Sans font (Latin, Cyrillic, Greek, Vietnamese, Devanagari). Other scripts (e.g. Arabic, Hebrew, CJK), or any renderer failure, fall back to Markdown for that video instead of failing the run. |
| `docx` | Uses `html2docx` and `python-docx`.                                                                                                                                                                                              |

## File layout

<Tree>
  <Tree.Folder name="output" defaultOpen>
    <Tree.File name="Video Title.md" />

    <Tree.File name="Video Title.html" />

    <Tree.File name="Video Title.docx" />

    <Tree.File name="Video Title_quiz.md" />

    <Tree.File name="Video Title_transcript.json" />

    <Tree.Folder name="Video Title" defaultOpen>
      <Tree.File name="01_Introduction.md" />

      <Tree.File name="02_Main Topic.md" />

      <Tree.File name=".notewise-output.json" />
    </Tree.Folder>
  </Tree.Folder>
</Tree>

Filenames are sanitized and capped at 100 characters. Batch collisions may append the video ID.