AI-Parrot Loaders
ai-parrot-loaders provides document loaders for AI-Parrot RAG (Retrieval-Augmented Generation) pipelines. Each loader transforms a specific document format into text chunks that can be embedded and searched.
Installation
pip install ai-parrot-loaders
Install only the extras you need:
pip install ai-parrot-loaders[pdf]
pip install ai-parrot-loaders[youtube]
pip install ai-parrot-loaders[audio]
pip install ai-parrot-loaders[web]
# Everything
pip install ai-parrot-loaders[all]
Available Extras
| Extra | Description |
|---|---|
pdf |
PDF loading with OCR support (PaddleOCR) |
youtube |
YouTube transcript and video download |
audio |
Audio transcription (WhisperX, pyannote) |
web |
HTML/web page loading |
ebook |
Structured EPUB and MOBI e-book loading |
documents |
PDF, DOCX, EPUB, and MOBI ingestion without OCR dependencies |
video |
Video processing (MoviePy, FFmpeg) |
Supported Formats
uv pip install 'ai-parrot[wiki]' automatically installs
ai-parrot-loaders[documents] and the parsing dependencies for PDF, DOCX, EPUB,
and MOBI. To use those loaders independently, install
uv pip install 'ai-parrot-loaders[documents]'.
The documents extra uses the native PDF and DOCX loaders plus the ebook
extra. The existing document extra is separate and supplies Mammoth for
document conversion; pdf includes the broader OCR stack.
| Loader | Format | Description |
|---|---|---|
TextLoader |
.txt |
Plain text files |
CSVLoader |
.csv |
CSV files |
ExcelLoader |
.xlsx, .xls |
Excel spreadsheets |
MSWordLoader |
.docx |
Microsoft Word documents |
HTMLLoader |
.html |
HTML files |
MarkdownLoader |
.md |
Markdown files |
PDFLoader |
.pdf |
PDF documents |
PDFMarkdownLoader |
.pdf |
PDF to Markdown conversion |
PDFTablesLoader |
.pdf |
PDF table extraction |
PowerPointLoader |
.pptx |
PowerPoint presentations |
EpubLoader |
.epub |
EPUB e-books |
MobiLoader |
.mobi |
MOBI/KF8 e-books via KindleUnpack |
WebLoader |
URL | Web pages |
YoutubeLoader |
URL | YouTube video transcripts |
VimeoLoader |
URL | Vimeo video transcripts |
AudioLoader |
.mp3, .wav, etc. |
Audio transcription |
VideoLoader |
URL | Video download + transcription |
VideoLocalLoader |
.mp4, etc. |
Local video transcription |
DocumentConverterLoader |
multiple | Auto-detect format and convert |
Quick Start
Structured EPUB and MOBI books
Install the optional ebook dependencies:
uv pip install 'ai-parrot-loaders[ebook]'
This includes ebooklib, beautifulsoup4, markdownify, and mobi (KindleUnpack).
from parrot_loaders.epubloader import EpubLoader
from parrot_loaders.mobiloader import MobiLoader
from parrot.loaders.ebook import ebook_sections
documents = await EpubLoader("book.epub").load()
sections = ebook_sections(documents)
await pageindex.insert_ebook("my-book", [section.model_dump() for section in sections])
# Same output contract and loader options for MOBI:
mobi_documents = await MobiLoader("book.mobi").load()
Create the PageIndex tree before inserting. Bookstore, GraphIndex and wiki ingestion select this structural path automatically for EPUB and MOBI documents.
bookstore add book.mobi --no-llm
MOBI support handles unencrypted MOBI7 HTML/NCX and KF8 EPUB output. MOBI7 uses
NCX navigation when present, then explicit nested HTML navigation, then HTML
heading structure. If the source has no hierarchy, the loader cannot recover
one that was never encoded. Invalid/encrypted books and PDF-only Print Replica
output raise MobiLoaderError. Decoding runs off the event loop with serialized
backend calls, and temporary files are removed on both success and failure.
Section provenance always refers to the original .mobi source.
Each document contains one section's body and an ebook_section metadata record:
section_id, parent_id, title, href (including fragments), depth,
toc_order, reading_order, source_uri, target_found, and origin.
per_chapter=False returns one readable book document with complete records in
ebook_sections. include_toc_document=True adds a nested navigation document.
Navigation parents and short TOC sections are retained even when they have no
body. Missing targets remain in the index with target_found=False. When a
document has no TOC entries, HTML headings supply its fallback structure.
Reading order follows the spine, while TOC order is recorded separately.
Repeated targets retain distinct navigation identities; their body is stored
once, on the last entry for that target, rather than duplicated.
Compatibility: load() defaults to split_documents=False for ebooks.
Explicit chunking remains available for vector-store callers. Section bodies no
longer carry synthetic Section:/Title: prefixes or duplicate chapter headings;
titles and relationships are metadata. Use ebook_markdown(sections) for a
readable Markdown export. The structural PageIndex path preserves all navigation
levels and does not merge short nodes or infer parents from Markdown headings.
Other documents
from parrot_loaders.factory import get_loader_class
# Auto-detect loader by file extension
LoaderClass = get_loader_class("report.pdf")
loader = LoaderClass(source="report.pdf")
documents = await loader.load()
for doc in documents:
print(doc.page_content[:200])
Or use a specific loader directly:
from parrot_loaders.youtube import YoutubeLoader
loader = YoutubeLoader(source="https://www.youtube.com/watch?v=...")
documents = await loader.load()
Requirements
- Python >= 3.11
- ai-parrot >= 0.23.18
License
MIT
1 maintainer
Metadata
1 maintainer
Release files for ai-parrot-loaders 1.0.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| ai_parrot_loaders-1.0.0.tar.gz | 148.4 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| ai_parrot_loaders-1.0.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 294.8 kB
Release files / ai_parrot_loaders-1.0.0.tar.gz
| Download URL | ai_parrot_loaders-1.0.0.tar.gz |
|---|---|
| Size | 148.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
235d0d8deb089a267a7cafcf19c0467ebbd6975cae73723b635632e6f36a5d24
|
|
BLAKE2b-256 checksum How to use checksums |
065c5da3d24ac3e22c7e68fe1ad39f51a1b0b0e6f554dc603f829cd2b92a7ca6
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 6, 2026.
Transparency logRelease files / ai_parrot_loaders-1.0.0-py3-none-any.whl
| Download URL | ai_parrot_loaders-1.0.0-py3-none-any.whl |
|---|---|
| Size | 146.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
3302156e78ae6ebdcb1d0734b8baa6b1627429bcc9ce57709414295498ec78bf
|
|
BLAKE2b-256 checksum How to use checksums |
e2251a869ca7e64faeae58a6c68dc2646687b917f3a2826c3263919fa6c4d5ac
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 6, 2026.
Transparency log
