Attachments: The Python Funnel for LLM Context and Multimodal Data

Introduction

attachments is a powerful Python library designed to simplify providing context to Large Language Models (LLMs). It acts as a universal funnel, transforming various file types, including PDFs, images, web pages, and even entire code repositories, into model-ready text and base64 encoded images. With attachments, developers can integrate diverse data sources into their LLM applications with just a few lines of code, streamlining multimodal AI development.

Installation

Getting started with attachments is straightforward. You can install it using pip:

pip install attachments

For advanced features like CSS selector highlighting or Microsoft Office support, specific extras can be installed:

# For CSS selector highlighting (requires Playwright)
pip install attachments[browser]
playwright install chromium

# For Microsoft Office format support
pip install attachments[office]

Examples

attachments offers a simple API for common use cases and a powerful Domain Specific Language (DSL) for advanced processing.

Quick Start

Transform any file into LLM-ready content:

from attachments import Attachments

# Process a local file or a URL
ctx = Attachments("path/to/file.pdf") 
# or 
ctx = Attachments("https://example.com/document.docx")

llm_ready_text   = str(ctx)       # All extracted text, "prompt-engineered"
llm_ready_images = ctx.images     # list[str] – base64 PNGs

print(f"Extracted text length: {len(llm_ready_text)}")
print(f"Number of images: {len(llm_ready_images)}")

LLM Integration

attachments provides direct integration with popular LLM APIs like OpenAI and Anthropic, formatting the content appropriately:

from openai import OpenAI
from attachments import Attachments

# Process a PowerPoint presentation, selecting slides 3 to 5
pptx = Attachments("https://github.com/MaximeRivest/attachments/raw/main/src/attachments/data/sample_multipage.pptx[3-5]")

client = OpenAI()
resp = client.chat.completions.create(
    model="gpt-4.1-nano", # Use an appropriate vision-capable model
    messages=pptx.openai_chat("Analyze the following document:")
)
print(resp.choices[0].message.content)

Advanced DSL Usage

The DSL allows for precise control over content extraction and transformation:

• Select pages or slides: report.pdf[1,3-5,-1]
• Web content selection: url[select:title]
• Image transformations: photo.jpg[rotate:90][crop:100,100,400,300]
• Repository analysis: ./my-project[ignore:standard][max_files:100]

Why Use Attachments?

attachments stands out for several key reasons:

• Comprehensive Multimodal Support: It handles a wide array of formats out of the box, including PDFs, PowerPoint, Word, Excel, CSV, TXT, Markdown, HTML, various image types, ZIP archives, and even Git repositories.
• Simplified LLM Context: It abstracts away the complexity of parsing and formatting diverse data for LLMs, providing a unified text and images output.
• Powerful DSL: The intuitive Domain Specific Language allows for granular control over content extraction, filtering, and transformation, enabling complex workflows with concise syntax.
• Extensibility: The modular pipeline architecture allows users to easily extend its capabilities by adding custom loaders, modifiers, presenters, refiners, and adapters.
• Direct LLM API Integration: It provides helper methods to format content directly for OpenAI, Anthropic, and DSPy, saving development time.
• Advanced Features: Capabilities like CSS selector highlighting for web scraping and dedicated Microsoft Office support enhance its utility for specialized tasks.

Links

Explore the attachments project further on GitHub:

• GitHub Repository: MaximeRivest/attachments