Pipelines

End-to-end pipeline shapes. Coming soon. Planned worked examples:

• RAG ingest. Walk a corpus, chunk by heading, embed with source provenance, write to a vector store.
• Agent tool use. Wire udoc into an LLM agent loop with the agent instructions page as context.
• Document classification. Run a layout-detection hook, classify by region label, route by type.
• Forensic audit. Read tracked changes, comments, and form fields from a DOCX corpus.
• Spreadsheet ingest. Multi-sheet typed cells into Pandas / DuckDB / Parquet.

Until they land, the Library guide covers Corpus, text_chunks, and the streaming Extractor.