# Iyana/llama-fs: A self-organizing file system with llama 3 Source: https://github.com/iyaja/llama-fs ## Summary LlamaFS is a self-organizing file manager that automatically renames and organizes files based on their content and conventions such as time, supporting images through Moondream and audio through Whisper. It runs in a batch mode that suggests a file structure for a submitted directory, and a watch mode, an interactive daemon that learns from the user's recent edits to create and move files proactively. The project uses Llama 3 through Groq for cloud inference, with an incognito toggle that routes requests through Ollama running the same model for local processing. The developers report that most file operations take under 500ms in watch mode, a benchmark by AgentOps. ## Article LlamaFS Inspiration Watch the explainer video Open your ~/Downloads directory. Or your Desktop. It's probably a mess... There are only two hard things in Computer Science: cache invalidation and naming things. What it does LlamaFS is a self-organizing file manager. It automatically renames and organizes your files based on their content and well-known conventions (e.g., time). It supports many kinds of files, including images (through Moondream) and audio (through Whisper). LlamaFS runs in two "modes" - as a batch job (batch mode), and an interactive daemon (watch mode). In batch mode, you can send a directory to LlamaFS, and it will return a suggested file structure and organize your files. In watch mode, LlamaFS starts a daemon that watches your directory. It intercepts all filesystem operations and uses your most recent edits to proactively learn how you rename file. For example, if you create a folder for your 2023 tax documents, and start moving 1-3 files in it, LlamaFS will automatically create and move the files for you! Uh... Sending all my personal files to an API provider?! No thank you! It also has a toggle for "incognito mode," allowing you route every request through Ollama instead of Groq. Since they use the same Llama 3 model, the perform identically. How we built it We built LlamaFS on a Python backend, leveraging the Llama3 model through Groq for file content summarization and tree structuring. For local processing, we integrated Ollama running the same model to ensure privacy in incognito mode. The frontend is crafted with Electron, providing a sleek, user-friendly interface that allows users to interact with the suggested file structures before finalizing changes. It's extremely fast! (by LLM standards)! Most file operations are processed in <500ms in watch mode (benchmarked by AgentOps). This is because of our smart caching that selectively rewrites sections of the index based on the minimum necessary filesystem diff. And of course, Groq's super fast inference API. 😉 It's immediately useful - It's very low friction to use and addresses a problem almost everyone has. We started using it ourselves on this project (very Meta). What's next for LlamaFS Find and remove old/unused files We have some really cool ideas for - filesystem diffs are hard... Installation Prerequisites Before installing, ensure you have the following requirements: Python 3.10 or higher pip (Python package installer) Installing To install the project, follow these steps: Clone the repository: git clone https://github.com/iyaja/llama-fs.git Navigate to the project directory: Install requirements pip install -r requirements.txt Update your .env Copy .env.example into a new file called .env. Then, provide the following API keys: Groq: You can obtain one from here. AgentOps: You can obtain one from here. Groq is used for fast cloud inference but can be replaced with Ollama in the code directly (TODO.) AgentOps is used for logging and monitoring and will report the latency, cost per session, and give you a full session replay of each LlamaFS call. (Optional) Install moondream if you want to use the incognito mode Usage To serve the application locally using FastAPI, run the following command This will run the server by default on port 8000. The API can be queried using a curl command, and passing in the file path as the argument. For example, on the Downloads folder: curl -X POST http://127.0.0.1:8000/batch \ -H "Content-Type: application/json" \ -d '{"path": "/Users//Downloads/", "instruction": "string", "incognito": false}' ## Comments **timdaub.eth**: Funnily enough, I know the co-working space this demo was filmed in because major parts of Kiwi News Protocol were developed in it: https://x.com/AlexReibman/status/1789895425828204553 **macbudkowski.eth**: Since my Downloads folder is an absolute mess, gonna test it out. Do you know @freeatnet if I can specify whether I prefer to have multiple folders for each category (just like in the demo) or just have all files in one folder?