blazegraph-io

Semantic document graph extraction. Transform PDFs into structured, queryable graphs for RAG, search, and document understanding.

Features

• Semantic graph output — Preserves document structure (sections, paragraphs, lists, tables)
• Bounding boxes — Every node maps to exact PDF coordinates
• Hierarchical structure — Parent-child relationships preserved
• Fast — Native Rust with embedded Tika for PDF parsing
• Local-first — No API key required, runs entirely on your machine

Installation

Build from source

git clone https://github.com/AmplifyTechnology/blazegraph-io.git
cd blazegraph-io
cargo build --release -p blazegraph-cli

Run

# Parse a PDF to JSON graph
./target/release/blazegraph-cli -i document.pdf -o graph.json

# With custom config
./target/release/blazegraph-cli -i document.pdf -c config.yaml -o graph.json

# See all options
./target/release/blazegraph-cli --help

Note: On first run, the CLI will automatically download a Java Runtime (~60MB) for PDF processing. It's cached for future use:

• macOS/Linux: ~/.local/share/blazegraph/jre
• Windows: %LOCALAPPDATA%\blazegraph\jre

Output Formats

Format	Description
graph	Full graph structure with nodes and edges (default)
sequential	Ordered segments with hierarchy info (good for RAG)

# Sequential format for RAG pipelines
./target/release/blazegraph-cli -i document.pdf -f sequential -o chunks.json

Configuration

See blazegraph-cli/configs/processing/ for example configuration files. These control:

• Section detection thresholds
• List detection patterns
• Spatial clustering parameters
• Size enforcement (max chunk size)

Project Structure

blazegraph-io/
├── blazegraph-core/     # Core parsing library
├── blazegraph-cli/      # Command-line interface
└── Cargo.toml           # Workspace definition

License

Licensed under either of:

• Apache License, Version 2.0 (LICENSE-APACHE)
• MIT license (LICENSE-MIT)

at your option.

Roadmap

This project is actively developed. Here's what's planned:

Distribution

• Publish CLI to crates.io (blazegraph-io)
• Publish core library to crates.io (blazegraph-io-core)
• Publish Python wrapper to PyPI (blazegraph-io)

File Format Support

• PDF
• Markdown (.md)
• Word documents (.docx)

Schema & Output

• Stable v1 schema specification
• Schema documentation with examples
• Migration guide for schema changes

Documentation

• Getting started guide
• Configuration reference
• Integration examples (LangChain, LlamaIndex, etc.)
• Output schema reference

Contributions and feedback welcome! Open an issue to discuss.