Talk to Your Mac and Query Your Docs with On-Device AI
Ever wish you could just ask your computer a question about your own documents without uploading everything to the cloud? Or maybe you're tired of typing out long search queries when a quick spoken question would do. That's the itch this new project scratches.
It brings on-device voice AI right to your macOS terminal, letting you interact with your local files using your voice. No API keys, no data leaving your machine—just a direct conversation with your own documents.
What It Does
RCLI (RunAnywhere CLI) is a command-line tool for macOS that uses local AI models to process your voice and your documents. You speak a question or command into your microphone, and it transcribes, understands, and executes it against a specified folder of documents—all offline. It’s essentially a private, voice-powered search and Q&A engine for your local files.
Why It's Cool
The "on-device" part is the real game-changer here. Privacy is baked in; your documents and voice recordings never hit a remote server. This makes it perfect for sensitive codebases, internal documents, or personal notes.
It's also incredibly straightforward. You point it at a directory, and you can start asking things like "What's the main architecture discussed in the Q3 planning doc?" or "Summarize the key points from last week's meeting notes." The tool uses local embeddings and a local LLM to find relevant snippets and generate answers directly from your text files, PDFs, and other documents.
The implementation is cleverly modular, using established local AI tools under the hood, which means it's both powerful and transparent for developers who might want to peek at how it works or tweak it.
How to Try It
Getting started is pretty standard for a CLI tool. You'll need macOS and Python installed.
- Clone the repo:
git clone https://github.com/RunanywhereAI/RCLI.git cd RCLI - Install the package:
pip install -e . - Run the setup to configure your document path and models:
rcli setup - Start a voice session with your documents:
rcli voice --path /path/to/your/documents
Then just start talking. The repo has more detailed instructions on model options and configuration if you need to fine-tune things.
Final Thoughts
As a developer, I see immediate use for quickly navigating a large legacy codebase or querying my own scattered project notes without breaking flow. It feels like a natural step towards more intuitive developer tools. While the accuracy will depend on your local models and the quality of your documents, the core idea—private, voice-driven interaction with your own data—is executed well here. It's a solid foundation that makes on-device AI feel practical and immediately useful.
@githubprojects
Repository: https://github.com/RunanywhereAI/RCLI