Skip to content

bibr 🦫

bibliography rodent chews through scientific papers and returns structured JSON. Read PDF, DOCX, JATS XML, HTML, and ePub; extract metadata, references, sections, sentences, tables, figures, and equations.

Install bibr from PyPI to use the features documented here. See installation for Python and system requirements, optional extras, and contributor source installs.

uv init --python 3.12 paper-extraction
cd paper-extraction
uv add bibr
uv run bibr setup
uv run bibr chew paper.pdf -o paper.json

The setup wizard detects your hardware and configures OCR and metadata extraction. The first run may download models and additional runtimes.

Get started JSON schema v11.0 Explore Scienceverse ↗

Alpha: check important extractions against the source

Extraction quality varies by paper, language, layout, and model. Current evaluation is strongest for English social science papers. A valid JSON document can still contain mistakes. Read the known limitations and evaluation guide before relying on a field.

  • Native parsing and OCR

    Structured formats are parsed natively. PDFs combine layout detection, usable native text, and OCR where needed. Follow the pipeline.

  • Choose the work to run

    References use local parsing by default. Choose --refs llm for LLM parsing or --refs off to skip them. --no-llm disables downstream LLM extraction; PDF OCR is configured separately. Configure a run.

  • Inspect the evidence

    Sentence IDs, page links, processing warnings, and extraction provenance help you review results. Coverage depends on the input and stage. Explore the output.

  • Local or hosted

    Use local or cloud models from the CLI and Python, or deploy an HTTP API with bibr serve. Fully offline operation also requires local model assets and external enrichment to be disabled. Deploy bibr.

Use it your way

You want to… Go to
Process your first paper Quickstart
Choose dependencies for your hardware Installation / Tester guide
Work in Python or a notebook Python library
Connect an agent MCP server
Run an HTTP service Deployment / REST API
Look up a flag, setting, or JSON field CLI / Settings / Schema
Understand model use and accuracy limits LLM use / Known limitations
Contribute or measure extraction quality Development / Evaluation

The CLI, settings, and schema references are generated from the code when this site is built. Schema validation checks the output's structure; it does not establish that an extracted fact is correct.

bibr was originally built as a preprocessing backend for Metacheck and can be used independently.