I’m excited to share llm-fuse, my first CLI open-source project! This tool helps you quickly aggregate relevant files from a repository into a single text file (or multiple files if chunking is enabled), making it easy to provide LLMs with contextual information from codebases.
Motivation While working with LLMs, I found that getting them to understand a codebase required manually copying and pasting multiple files. I wanted a simple way to automate this—so I built llm-fuse (with over 90% of the code generated using ChatGPT o3-mini-high).
Features Scan a local directory (or only Git-tracked files) Clone and process remote Git repositories Include/exclude files using regex filters Estimate token counts Automatically chunk large files Generates an output file with a summary, file system diagram, and file contents Feedback Welcome! The tool has been tested on macOS, but I haven’t had the chance to test it on Windows yet—so any feedback is highly appreciated!
Next Steps I’m considering expanding llm-fuse to allow users to specify a starting file (e.g., UserService.java), and the tool would then automatically gather related classes and dependencies, ensuring LLMs have just enough context to answer questions with minimal hallucination.
Would love to hear your thoughts.
