Document Loaders
On this page
document_loaders is the document-ingestion module of the Zerda ecosystem. It converts files, web pages, directories, knowledge graphs, and tables into a single normalized unit — LoadedDocument — that the rest of the Fennec / RAG pipeline (chunking, embedding, indexing) can consume without caring what the original source format was.
This documentation is generated directly from static analysis of the
document_loaderssource package. Every signature, parameter, exception, and behavior described here is taken from the actual implementation. Where something cannot be determined from the code (e.g. version history, since no changelog or git metadata ships with this package), it is stated explicitly rather than guessed.
Why this module exists
RAG and LLM applications need text from many different sources — PDFs, Word docs, spreadsheets, scraped web pages, JSON exports, knowledge graphs. Each format has its own quirks (encodings, password protection, broken .doc files, rate-limited APIs, malformed CSV). document_loaders centralizes all of that handling behind one consistent contract:
loader = SomeLoader(...)
docs: list[LoadedDocument] = loader.load()
Every loader, regardless of source type, returns a list of LoadedDocument objects with page_content (the text) and metadata (a dict describing where it came from).
Module map
Where to go next
- New to the library? Start with Installation and Quickstart.
- Want the big picture first? Read Architecture.
- Looking for a specific class or method? Go to the API Reference Overview.
- Need recipes for real pipelines? See the Cookbook.
- Something broke? Check Troubleshooting.
Bilingual codebase
The source code mixes English docstrings with Arabic comments and docstrings (notably in graph_loader.py and table_loader.py), reflecting the project's focus on first-class Arabic/RTL document support. This documentation is written in English throughout, but Arabic text in code examples (e.g. sample node labels, CSV rows) is preserved as it appears in the source.