Document Extractor

A production-ready FastAPI service for extracting text content from documents (research papers, manuscripts). Designed to be called by n8n workflows.

Supported Formats

Format	Engine	Extension(s)
PDF	pymupdf4llm	.pdf
DOCX	python-docx + mammoth (fallback)	.docx
LaTeX	TexSoup + regex (fallback)	.tex
RTF	striprtf	.rtf
Plain Text	built-in decode	.md, .markdown, .txt, .text

Quick Start

# Build and run
docker compose up --build

# Health check
curl http://localhost:8000/health

API Endpoints

GET /health

Returns {"status": "ok"}.

POST /extract/urls

Extract text from documents at given URLs.

curl -X POST http://localhost:8000/extract/urls \
  -H "Content-Type: application/json" \
  -d '{"urls": ["https://example.com/paper.pdf"]}'

Response:

[
  {
    "url": "https://example.com/paper.pdf",
    "content": "# Title\n\nExtracted text...",
    "error": null
  }
]

POST /extract/binary

Extract text from uploaded binary files (designed for n8n integration).

curl -X POST http://localhost:8000/extract/binary \
  -F 'data=[{"binaryKey": "file", "fileName": "paper.pdf"}]' \
  -F 'file=@paper.pdf'

Response:

[
  {
    "binaryKey": "file",
    "fileName": "paper.pdf",
    "text": "Extracted text...",
    "originalBinaryKey": "file",
    "mdBinaryKey": "file_0",
    "mdFileName": "full.md"
  }
]

Configuration

All settings are configurable via environment variables:

Variable	Default	Description
MAX_WORKERS	cpu_count * 2	ThreadPoolExecutor size
MAX_CONCURRENT_DOWNLOADS	10	Max concurrent URL downloads
DOWNLOAD_TIMEOUT_SECONDS	30	HTTP download timeout
MAX_FILE_SIZE_MB	50	Maximum file size allowed
REQUEST_TIMEOUT_SECONDS	120	Overall request timeout
LOG_LEVEL	INFO	Logging level

Deployment on Dokploy

1. Push this repository to your Git provider
2. In Dokploy, create a new service pointing to the repository
3. Dokploy will automatically detect the docker-compose.yml and build/deploy
4. Set environment variables in Dokploy's UI as needed

Architecture

• No files stored on server — all processing is in-memory (PDF uses temp files that are immediately deleted)
• No ML/GPU — pure code-based extraction
• Concurrent processing — uses asyncio.gather with thread pools for CPU-bound extraction
• Graceful error handling — per-item errors never crash the batch; HTTP 200 always returned