chunkwrap 1.1.1

A project that splits files and copies them to paste buffer with context for an LLM.

chunkwrap

A Python utility for splitting large files into manageable chunks, masking secrets, and wrapping each chunk with custom prompts for Large Language Model (LLM) processing.

Overview

chunkwrap helps you prepare large files for LLM workflows by:

• Splitting them into smaller, prompt-ready chunks

• Redacting secrets via TruffleHog-style regexes

• Tracking progress across invocations

• Supporting clipboard-based interaction or (soon) alternate output modes

Features

• Configurable chunking: Choose chunk size (default: 10,000 characters)

• Multi-file support: Concatenate multiple inputs into a single stream

• Secret masking: Redact sensitive patterns using configurable regexes

• Prompt wrapping: Use distinct prompts for intermediate and final chunks

• Clipboard integration: Copy output chunk directly to your paste buffer

• State tracking: Progress is remembered across runs using a local .chunkwrap_state file

• Optional prompt suffix: Append boilerplate only to intermediate chunks

Installation

1. To install from source, clone the repository:

   git clone https://github.com/magicalbob/chunkwrap.git
   cd chunkwrap
   

   Or just install from PyPI:

   pip install chunkwrap
   

2. Install dependencies (if installed from source):

   pip install -e .
   

   Or for developer tools:

   pip install -e ".[dev]"
   

3. On first run, a default config file will be created at:

   • Linux/macOS: ~/.config/chunkwrap/config.json

   • Windows: %APPDATA%\chunkwrap\config.json

Usage

Minimal example

chunkwrap --prompt "Analyze this:" --file myscript.py

Multiple files

chunkwrap --prompt "Review each file:" --file a.py b.md

Secret masking

Place a truffleHogRegexes.json file in the same directory:

json

{
  "AWS": "AKIA[0-9A-Z]{16}",
  "Slack": "xox[baprs]-[0-9a-zA-Z]{10,48}"
}

Each match will be replaced with ***MASKED-<KEY>***.

Custom chunk size

chunkwrap --prompt "Summarize section:" --file notes.txt --size 5000

Final chunk prompt

chunkwrap --prompt "Analyze chunk:" --lastprompt "Now summarize everything:" --file long.txt

Disable prompt suffix

chunkwrap --prompt "Chunk:" --file script.py --no-suffix

Show config path

chunkwrap --config-path

Reset state

chunkwrap --reset

Output Format

Each chunk is wrapped like:

Your prompt (chunk 2 of 4)
"""
[redacted content]
"""

Final chunk omits the index and uses --lastprompt if provided.

Configuration File

On first run, chunkwrap creates a configuration file at the following path:

• Linux/macOS: ~/.config/chunkwrap/config.json

• Windows: %APPDATA%\chunkwrap\config.json

This file allows you to customize the default behavior of the tool. You can edit it manually to override any of the options below.

Available Options

{
  "default_chunk_size": 10000,
  "intermediate_chunk_suffix": " Please provide only a brief acknowledgment that you've received this chunk. Save your detailed analysis for the final chunk.",
  "final_chunk_suffix": "Please now provide your full, considered response to all previous chunks."
}

• default_chunk_size: (integer)
  Sets the default number of characters per chunk when --size is not specified on the command line.

• intermediate_chunk_suffix: (string)
  This text is appended to the --prompt on all intermediate (non-final) chunks unless the --no-suffix flag is used.

• final_chunk_suffix: (string)
  This text is appended to the --lastprompt (or --prompt, if --lastprompt is not used) for the final chunk. It's intended to signal to the LLM that a full, detailed response is now appropriate.

Example

You might modify your config to create tighter chunking and less verbose suffixes:

{
  "default_chunk_size": 5000,
  "intermediate_chunk_suffix": "Brief reply only, please.",
  "final_chunk_suffix": "Full summary now."
}

These values will be automatically merged with any defaults added in future releases, so missing keys will not cause errors.

Roadmap

Near-term improvements

1. Make cross platform: local usage on Mac is good. My test machine is via ssh to a linux machine. The current code does not support this. Consider adding optional argument chunkwrap [--output {clipboard|stdout|file}] to handle this situation.

Future considerations

• Chunk overlap: Add optional overlap between chunks to preserve context across boundaries
• Smart chunking: Break at natural boundaries (sentences, paragraphs) rather than arbitrary character counts
• Output formats: Support for different wrapper formats (XML tags, markdown blocks, etc.)
• Parallel processing: For very large file sets, allow processing multiple chunks simultaneously
• Integration modes: Direct API integration with popular LLM services

Requirements

• Python 3.11+

• pyperclip

License

GNU General Public License v3.0 --- see LICENSE for details.