Convert PDFs to Text Locally with Llama-Scan and Ollama

What you’ll get

You will convert any PDF document (including those with images and diagrams) into a plain text file on your own computer. This approach uses local AI models, meaning you don’t need an internet connection after the initial setup, and there are no ongoing costs or ‘token’ fees.

Tools you need

• Python (free): The programming language needed to run the llama-scan tool.
• Ollama (free): A tool that lets you run large language models (LLMs) directly on your computer.
• llama-scan (free): A Python tool that uses Ollama to read and convert PDFs into text.

Steps

1. Install Python: First, you need Python 3.10 or newer. If you don’t have it, download and install it from the official Python website (python.org). On Windows, make sure to check the box that says “Add Python to PATH” during installation.

2. Install Ollama and download the model: Ollama is a tool that lets you run powerful AI models directly on your computer. Go to the Ollama website (ollama.com) and download the installer for your operating system (Windows, macOS, or Linux). Follow the installation instructions. Once installed, Ollama will run in the background.

   Next, open your computer’s terminal (on Windows, search for ‘Command Prompt’ or ‘PowerShell’; on macOS/Linux, search for ‘Terminal’). Type this command and press Enter to download the qwen2.5vl model. This model can understand both text and images.

   ollama run qwen2.5vl:latest

   You will see a download progress. This might take some time depending on your internet speed. Once it’s done, the model will be ready to use. You can then type /bye and press Enter to exit the chat, or simply close the terminal window.

3. Install llama-scan: In the same terminal window (or a new one), type this command to install the llama-scan tool:

   pip install llama-scan

   You should see messages about llama-scan and its dependencies being installed. If it finishes without errors, it’s ready.

4. Prepare your PDF file: Make sure you have the PDF file you want to convert saved on your computer. Note down its full path (e.g., C:\Users\YourName\Documents\my_report.pdf on Windows, or /Users/YourName/Documents/my_report.pdf on macOS/Linux). Remember to use double quotes around paths that contain spaces.

5. Convert the PDF to text: In your terminal, use the llama-scan command followed by the full path to your PDF file. For example:

   # Windows (Command Prompt or PowerShell)
   llama-scan "C:\Users\YourName\Documents\my_report.pdf"

   # macOS or Linux
   llama-scan "/Users/YourName/Documents/my_report.pdf"

   By default, llama-scan will create a new folder called output in the same location where you run the command. Inside this folder, you will find a text file (.txt) containing the content of your PDF.

   Optional: Specify an output folder: If you want the text file to be saved somewhere else, use the --output option:

   # Windows (Command Prompt or PowerShell)
   llama-scan "C:\Users\YourName\Documents\my_report.pdf" --output "C:\MyTextFiles"

   # macOS or Linux
   llama-scan "/Users/YourName/Documents/my_report.pdf" --output "/home/YourName/MyTextFiles"

   Optional: Convert specific pages: To convert only a part of the PDF, use --start and --end:

   # Example: Convert pages 1 to 5
   llama-scan "document.pdf" --start 1 --end 5

   You will see messages in the terminal as llama-scan processes each page. Once it’s done, check your specified output folder (or the default output folder) for the new text file.

Original source

This workflow was created by nawazgafar and shared on Hackernews. It uses the open-source llama-scan tool to process PDFs with local AI models.

Notes & variations

• Common mistake: The most common issue is not having Ollama running in the background or not having the qwen2.5vl:latest model downloaded. Make sure Ollama is installed and the model download step is completed successfully before running llama-scan.
• Tip for better results: For complex PDFs, you can provide custom instructions to the model. Create a text file (e.g., instructions.txt) with specific guidance (e.g., “Summarize each section” or “Extract only the financial figures”). Then, use the --custom-instructions option: llama-scan your_file.pdf --custom-instructions instructions.txt.
• Alternative models: The llama-scan tool allows you to use other multimodal models supported by Ollama. You can find a list of available models on the Ollama website. To use a different model, replace qwen2.5vl:latest with your chosen model name in the ollama run command and the llama-scan command (using the --model option).