DocVert is an intelligent, LLM-powered CLI tool and agent for converting DOCX and PDF documents into clean, semantic Markdown. It focuses heavily on preserving document structure, headings, lists, and visual elements, while extracting key document metadata into sidecar JSON files.
- Robust DOCX Parsing: Primary parsing using
python-docxwith heuristic heading detection. Fallback tomammothfor difficult layouts. - Advanced PDF Parsing: High-fidelity PDF extraction using
doclingas the primary engine. Fallback tounstructuredfor edge cases. - Rich Output Format:
- Generates clean, semantic
.mdfiles. - Produces a sidecar
.jsonfile containing extraction confidence scores, metadata, and parsing warnings. - Automatically extracts and saves images referenced in the source documents.
- Generates clean, semantic
- Batch Processing & Caching: Efficiently process large directories of files with built-in caching to avoid redundant parsing.
- Provider-Agnostic LLM Refinement: Uses
litellmunder the hood, natively supporting OpenAI, Vertex AI, Anthropic, Bedrock, and local models via Ollama. - Air-Gapped / Offline Deployment: Pre-built Docker images via GitHub Releases for secure, offline environments.
- Developer Ready:
- 100% test coverage.
- Robust type hints powered by
pydantic. - Built-in CLI using modern Python tooling (
uv).
# Build from source
git clone https://github.com/seonghobae/docvert.git
cd docvert
docker build -t docvert:offline .
# Convert a file
docker run --rm -v $(pwd):/data \
docvert:offline convert /data/input.pdf --output-dir /data/outOr install natively — see the full Installation Guide.
Full documentation is available at seonghobae.github.io/docvert
- Installation Guide (English) — Per-OS zero-setup, no Homebrew required
- 설치 가이드 (한국어) — OS별 제로셋업 설치 안내
- User Manual & CLI Reference — CLI usage, LLM configuration, architecture
- Offline Deployment Runbook — Air-gapped setup guide
- Architecture Decision Records — Parser implementation choices
Pre-built Docker images for air-gapped environments:
github.com/seonghobae/docvert/releases
MIT License