Skip to content

Installation

Requirements

  • Python 3.12+
  • An AI provider — one of:
    • AWS account with Bedrock model access in us-east-1 (Nova 2 Lite + Nova 2 multimodal embeddings) — the default, or
    • an OpenAI API key (llm_provider: openai runs the whole pipeline on GPT-5 mini + text-embedding-3; no AWS needed at all).
  • A vector database — or none: the sqlite connector stores vectors in a local file with zero setup. The other seven (Pinecone, Qdrant, pgvector, MongoDB, Milvus, OpenSearch, Weaviate) each need one connection URL or API key. See Vector stores.
  • A reranker — a free Jina key (default), Amazon Rerank (reranker: aws, same AWS credentials), or reranker: none.
  • A GPU for OCR — Apple Silicon (Metal) or NVIDIA. Parse runs a local vision-language model behind an inference server; everything else is CPU-fine.

Install the package

pip install ingestlib          # or: uv add ingestlib

From source:

git clone https://github.com/LangModule/ingestlib.git
cd ingestlib
uv sync

System dependency: LibreOffice

Only needed for DOCX/PPTX input (they are converted to PDF first):

brew install --cask libreoffice          # macOS (installs the `soffice` binary)
sudo apt install libreoffice-core libreoffice-writer libreoffice-impress   # Linux

PDFs work without it.

Start the OCR inference server

Parse runs PaddleOCR-VL-1.6 (a 0.9B vision-language model) behind a local inference server. The first launch downloads ~1.8 GB of weights; later launches load from cache in seconds.

uv run python -m mlx_vlm.server --port 8111 --model PaddlePaddle/PaddleOCR-VL-1.6

vllm serve PaddlePaddle/PaddleOCR-VL-1.6 --port 8111

Then set paddle_vl.backend: vllm-server in config.yaml.

The layout model (PP-DocLayoutV3, ~126 MB) auto-downloads on the first parse and runs on CPU.

Which operations need the server?

Only parse (and therefore ingest). classify and split in standalone mode read the PDF's native text directly, and retrieve never touches OCR.

Disk footprint

Component Size Location
Python dependencies ~3 GB your virtualenv
PaddleOCR-VL-1.6 weights ~1.8 GB ~/.cache/huggingface/hub/
PP-DocLayoutV3 ~126 MB ~/.paddlex/official_models/
LibreOffice ~600 MB system

Next: Configuration.