ingest-cli 1.0.2

High-quality document processing for RAG pipelines, supporting multiple formats and processing backends

ingest

High-quality document processing CLI for RAG pipelines. Process PDFs, Office documents, images, and more into markdown, JSON, HTML, or RAG-optimized chunks.

Features

✅ Standalone CLI Command - Simple ingest command
✅ 4 Output Formats - markdown, json, html, chunks (RAG-optimized)
✅ 3 Converters - pdf, table, ocr specialized processing
✅ LLM Enhancement - Optional AI boost (81% → 91% table accuracy)
✅ Multi-Worker - Parallel batch processing
✅ 20+ Options - Full control over processing

Quick Start

Installation

Using uv (recommended - fastest!)

# Basic installation (fast, lightweight)
uv pip install ingest-cli

# With marker-pdf for high-quality processing
uv pip install ingest-cli[marker]

# With LLM support
uv pip install ingest-cli[llm]

# Full installation (everything)
uv pip install ingest-cli[full]

Using pip

# Basic installation
pip install ingest-cli

# With marker-pdf
pip install ingest-cli[marker]

# Full installation
pip install ingest-cli[full]

Install from source

git clone https://github.com/therealtimex/ingest.git
cd ingest

# Basic installation (lightweight, no marker-pdf)
uv pip install -e .

# With marker-pdf for high-quality processing
uv pip install -e ".[marker]"

# Full installation with all features
uv pip install -e ".[full]"

Basic Usage

# Process a document
ingest document.pdf

# Process for RAG
ingest ./documents --output-format chunks --batch-mode

# Extract tables with LLM
ingest report.pdf --converter-type table --use-llm

# View help
ingest --help

Common Use Cases

1. RAG System Preparation

ingest ./knowledge_base \
    --output-format chunks \
    --batch-mode \
    --workers 4

Output: Pre-chunked JSON optimized for embeddings and retrieval.

2. Table Extraction

ingest financial_reports/ \
    --converter-type table \
    --use-llm \
    --output-format json \
    --batch-mode

Output: High-accuracy table data in JSON format.

3. OCR Scanned Documents

ingest scanned_docs/ \
    --force-ocr \
    --output-format markdown \
    --batch-mode

Output: Clean markdown from scanned PDFs.

Output Formats

• markdown: Clean markdown with proper formatting
• json: Structured JSON with full metadata
• html: Web-ready HTML with embedded images
• chunks: RAG-optimized pre-chunked JSON for vector databases

Performance

Workers	VRAM	Throughput (H100)
1	5GB	~30 pages/sec
4	20GB	~120 pages/sec
8	40GB	~240 pages/sec

Requirements

• Python 3.10+
• Optional: GPU for faster processing (CPU mode available)

Environment Variables

# PyTorch device
export TORCH_DEVICE=cuda  # or cpu, mps

# LLM API keys (optional, for enhanced accuracy)
export GOOGLE_API_KEY="your-gemini-key"
export ANTHROPIC_API_KEY="your-claude-key"
export OPENAI_API_KEY="your-openai-key"

License

MIT License - see LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

• Issues: GitHub Issues
• Repository: github.com/therealtimex/ingest

---

Built with ❤️ by RealTimeX