Localizer

A tool for processing and chatting with files using local LLMs via Ollama. Available as both a CLI and a web interface. Supports text, code, PDFs, and images.

Features

• Process files - Read files, send to LLM with instructions, save output
• Chat with files - Interactive Q&A about file contents
• Text-to-speech - Read files aloud using macOS voices
• Voice recording - Record from microphone, transcribe with Whisper or save audio
• Reply assistant - Polish and refine professional message replies
• Web interface - Browser-based UI with streaming responses, drag-and-drop uploads, and microphone selector

Requirements

• macOS (uses native file pickers and say command)
• Ollama with at least one model installed
• Node.js 18+ (for the web interface)

Installation

1. Clone or download this repository
2. Make the CLI script executable:

   chmod +x localizer

3. Run the setup wizard:

   ./localizer setup

The setup wizard will:

• Check and install required dependencies (Ollama) via Homebrew
• Offer to install optional dependencies (sox, whisper, tesseract, poppler)
• Pull Ollama models if none are installed
• Add localizer to your PATH or create a shell alias

Manual Installation

If you prefer manual setup, install dependencies yourself:

Feature	Dependency	Install
LLM processing	Ollama (required)	brew install --cask ollama
PDF support	pdftotext	brew install poppler
Image OCR	tesseract	brew install tesseract
Native image understanding	Vision model	ollama pull llava
Audio transcription	whisper	brew install openai-whisper
Audio conversion	ffmpeg	brew install ffmpeg

Then pull at least one model:

ollama pull llama3.2

Web Interface

A browser-based UI that exposes all localizer features with a modern dark-themed design, real-time streaming, and drag-and-drop file uploads.

Quick Start

cd web
npm run install:all
npm run dev

This starts both the Express backend (port 3001) and the Vite frontend (port 5173). Open http://localhost:5173 in your browser.

Web Features

• Process - Upload files, pick a model, enter instructions, view streaming output
• Chat - Upload context files, multi-turn conversation with streaming responses
• Reply - Paste a message, write a draft, choose a tone, get a polished reply
• Speak - Paste text or upload a file, select voice and speed, trigger TTS
• Record - Browser-based recording with microphone selection, transcribe via Whisper
• Setup - View Ollama status, check dependencies, pull new models

Tech Stack

• Frontend: React 18, Vite, Tailwind CSS
• Backend: Node.js, Express
• LLM: Ollama local API (localhost:11434)
• Recording: Browser MediaRecorder API (no sox required)
• Transcription: Whisper via ffmpeg + openai-whisper

CLI Usage

./localizer <command> [path]

If no path is provided, a macOS file picker dialog will open.

Commands

process [path]

Read files, process with LLM instructions, and write output to a file.

./localizer process ./src # Process a folder
./localizer process ./document.pdf # Process a PDF
./localizer process # Opens file picker

chat [path]

Interactively chat and ask questions about file contents.

./localizer chat ./data # Chat about files in a folder
./localizer chat ./readme.md # Chat about a single file
./localizer chat # Opens file picker

speak [path]

Read file contents aloud using text-to-speech.

./localizer speak ./notes.txt # Speak a file
./localizer speak # Opens file picker

You can select from available macOS voices and adjust speech rate.

record

Record audio from microphone with options to transcribe or save.

./localizer record

Options:

1. Transcribe to text - Uses Whisper for speech-to-text
2. Save as audio file - Saves the recording as a .wav file

Press Ctrl+C to stop recording.

reply

Polish and refine professional message replies using local LLMs.

./localizer reply

Features:

• Paste the original message context
• Enter your draft reply or key points
• Choose from multiple tones (Professional, Friendly, Formal, Direct, Diplomatic)
• Copy polished result to clipboard
• Iterate with feedback until satisfied

setup

One-time setup wizard that configures everything needed to run localizer.

./localizer setup

The wizard walks you through 3 steps:

1. Dependencies - Checks for and installs required/optional tools via Homebrew
2. Ollama Models - Pulls recommended LLM models if none are installed
3. Shell Config - Adds localizer to your PATH or creates an alias

Supported File Types

Category	Extensions
Text	.txt, .md, .json, .yaml, .yml, .xml, .csv
Code	.js, .ts, .py, .go, .rs, .java, .c, .cpp, .h, .sh, .html, .css, .sql, .swift, .kt, .rb, .php
PDF	.pdf (requires pdftotext)
Images	.png, .jpg, .jpeg, .gif, .webp, .bmp, .tiff

Image Processing

Images can be processed in two ways:

1. Vision models (llava, bakllava, moondream, etc.) - Native image understanding
2. OCR fallback - Text extraction via Tesseract for non-vision models

Examples

# Process a codebase with instructions
./localizer process ./src
>"Summarize this codebase and list all API endpoints"# Chat about a PDF document
./localizer chat ~/Documents/report.pdf
You: What are the key findings?
Assistant: ...
# Have your notes read aloud
./localizer speak ~/notes.md
# Record a voice memo and transcribe it
./localizer record
# Write a polished professional reply
./localizer reply
# Start the web interfacecd web && npm run dev

License

MIT