pdf-file-renamer 0.4.2

Intelligent PDF renaming using LLMs

PDF Renamer

Intelligent PDF file renaming using LLMs. This tool analyzes PDF content and metadata to suggest descriptive, standardized filenames.

🚀 Works with OpenAI, Ollama, LM Studio, and any OpenAI-compatible API

Features

• Advanced PDF parsing using docling-parse for better structure-aware extraction
• OCR fallback for scanned PDFs with low text content
• Smart LLM prompting with multi-pass analysis for improved accuracy
• Suggests filenames in format: Author-Topic-Year.pdf
• Dry-run mode to preview changes before applying
• Enhanced interactive mode with options to accept, manually edit, retry, or skip each file
• Live progress display with concurrent processing for speed
• Configurable concurrency limits for API calls and PDF extraction
• Batch processing of multiple PDFs with optional output directory

Installation

Quick Start (No Installation Required)

# Run directly with uvx
uvx pdf-renamer --dry-run /path/to/pdfs

Install from PyPI

# Using pip
pip install pdf-file-renamer

# Using uv
uv pip install pdf-file-renamer

Install from Source

# Clone and install
git clone https://github.com/nostoslabs/pdf-renamer.git
cd pdf-renamer
uv sync

Configuration

Configure your LLM provider:

Option A: OpenAI (Cloud)

cp .env.example .env
# Edit .env and add your OPENAI_API_KEY

Option B: Ollama or other local models

# No API key needed for local models
# Either set LLM_BASE_URL in .env or use --url flag
echo "LLM_BASE_URL=http://patmos:11434/v1" > .env

Usage

Quick Start

# Preview renames (dry-run mode)
pdf-renamer --dry-run /path/to/pdf/directory

# Actually rename files
pdf-renamer --no-dry-run /path/to/pdf/directory

# Interactive mode - review each file
pdf-renamer --interactive --no-dry-run /path/to/pdf/directory

Using uvx (No Installation)

# Run directly without installing
uvx pdf-renamer --dry-run /path/to/pdfs

# Run from GitHub
uvx https://github.com/nostoslabs/pdf-renamer --dry-run /path/to/pdfs

Options

• --dry-run/--no-dry-run: Show suggestions without renaming (default: True)
• --interactive, -i: Interactive mode with rich options:
  • Accept - Use the suggested filename
  • Edit - Manually modify the filename
  • Retry - Ask the LLM to generate a new suggestion
  • Skip - Skip this file and move to the next
• --model: Model to use (default: llama3.2, works with any OpenAI-compatible API)
• --url: Custom base URL for OpenAI-compatible APIs (default: http://localhost:11434/v1)
• --pattern: Glob pattern for files (default: *.pdf)
• --output-dir, -o: Move renamed files to a different directory
• --max-concurrent-api: Maximum concurrent API calls (default: 3)
• --max-concurrent-pdf: Maximum concurrent PDF extractions (default: 10)

Examples

Using OpenAI:

# Preview all PDFs in current directory
uvx pdf-renamer --dry-run .

# Rename PDFs in specific directory
uvx pdf-renamer --no-dry-run ~/Documents/Papers

# Use a different OpenAI model
uvx pdf-renamer --model gpt-4o --dry-run .

Using Ollama (or other local models):

# Using Ollama on patmos server with gemma model
uvx pdf-renamer --url http://patmos:11434/v1 --model gemma3:latest --dry-run .

# Using local Ollama with qwen model
uvx pdf-renamer --url http://localhost:11434/v1 --model qwen2.5 --dry-run .

# Set URL in environment and just use model flag
export LLM_BASE_URL=http://patmos:11434/v1
uvx pdf-renamer --model gemma3:latest --dry-run .

Other examples:

# Process only specific files
uvx pdf-renamer --pattern "*2020*.pdf" --dry-run .

# Interactive mode with local model
uvx pdf-renamer --url http://patmos:11434/v1 --model gemma3:latest --interactive --no-dry-run .

# Run directly from GitHub
uvx https://github.com/nostoslabs/pdf-renamer --no-dry-run ~/Documents/Papers

Interactive Mode

When using --interactive mode, you'll be presented with each file one at a time with detailed options:

================================================================================
Original: 2024-research-paper.pdf
Suggested: Smith-Machine-Learning-Applications-2024.pdf
Confidence: high
Reasoning: Clear author and topic identified from abstract
================================================================================

Options:
  y / yes / Enter - Accept suggested name
  e / edit - Manually edit the filename
  r / retry - Ask LLM to generate a new suggestion
  n / no / skip - Skip this file

What would you like to do? [y]:

This mode is perfect for:

• Reviewing suggestions before applying them
• Fine-tuning filenames that are close but not quite right
• Retrying when the LLM suggestion isn't good enough
• Building confidence in the tool before batch processing

You can use interactive mode with --dry-run to preview without actually renaming files, or with --no-dry-run to apply changes immediately after confirmation.

How It Works

1. Extract: Uses docling-parse to read first 5 pages with structure-aware parsing, falls back to PyMuPDF if needed
2. OCR: Automatically applies OCR for scanned PDFs with minimal text
3. Metadata Enhancement: Extracts focused hints (years, emails, author sections) to supplement unreliable PDF metadata
4. Analyze: Sends full content excerpt to LLM with enhanced metadata and detailed extraction instructions
5. Multi-pass Review: Low-confidence results trigger a second analysis pass with focused prompts
6. Suggest: LLM returns filename in Author-Topic-Year format with confidence level and reasoning
7. Interactive Review (optional): User can accept, edit, retry, or skip each suggestion
8. Rename: Applies suggestions (if not in dry-run mode)

Cost Considerations

OpenAI:

• Uses gpt-4o-mini by default (very cost-effective)
• Processes first ~4500 characters per PDF
• Typical cost: ~$0.001-0.003 per PDF

Ollama/Local Models:

• Completely free (runs on your hardware)
• Works with any Ollama model (llama3, qwen2.5, mistral, etc.)
• Also compatible with LM Studio, vLLM, and other OpenAI-compatible endpoints

Filename Format

The tool generates filenames in this format:

• Smith-Kalman-Filtering-Applications-2020.pdf
• Adamy-Electronic-Warfare-Modeling-Techniques.pdf
• Blair-Monopulse-Processing-Unresolved-Targets.pdf

Guidelines:

• First author's last name
• 3-6 word topic description (prioritizes clarity over brevity)
• Year (if identifiable)
• Hyphens between words
• Target ~80 characters (can be longer if needed for clarity)