R CLI

██████╗        ██████╗██╗     ██╗
██╔══██╗      ██╔════╝██║     ██║
██████╔╝█████╗██║     ██║     ██║
██╔══██╗╚════╝██║     ██║     ██║
██║  ██║      ╚██████╗███████╗██║
╚═╝  ╚═╝       ╚═════╝╚══════╝╚═╝

What is R CLI?

R CLI is a tool orchestrator for local LLMs. It exposes 74 "skills" (PDF generation, SQL queries, git, docker, etc.) as structured function calls that any OpenAI-compatible model can invoke.

This is NOT an operating system. It's a Python CLI that sits between your local LLM (Ollama, LM Studio) and real system tools.

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   You       │────▶│   R CLI     │────▶│  Local LLM  │
│  (prompt)   │     │ (orchestrator)│    │  (Ollama)   │
└─────────────┘     └──────┬──────┘     └──────┬──────┘
                          │                    │
                          ▼                    │
                   ┌─────────────┐             │
                   │   Skills    │◀────────────┘
                   │ (74 tools)  │  function call
                   └─────────────┘

$ r chat "Create a PDF report about Python"
# LLM calls pdf.generate_pdf() -> creates python_report.pdf

$ r sql sales.csv "SELECT product, SUM(revenue) FROM data GROUP BY product"
# Runs actual SQL against CSV using DuckDB

$ r rag --add ./docs/ && r rag --query "how does auth work"
# ChromaDB vectors, semantic search across your docs

Why Structured Tools Instead of Terminal Access?

You could just give an LLM shell access. But structured function calling provides:

Example: When you ask "compress python files", the LLM doesn't generate bash. It calls:

{
  "tool": "archive.create_zip",
  "arguments": {
    "source_path": ".",
    "pattern": "*.py",
    "output": "python_files.zip"
  }
}

R CLI validates the arguments, executes the tool, and returns structured results.

Features

Installation

# Basic
pip install r-cli-ai

# With all features
pip install r-cli-ai[all]

# R OS Simulator (Textual TUI)
pip install r-cli-ai[simulator]

# Raspberry Pi (with GPIO)
pip install r-cli-ai[all-rpi]

Requirements

• Python 3.10+
• Ollama or LM Studio
• 8GB+ RAM (16GB+ recommended)

Quick Start

1. Start your LLM

# Ollama
ollama pull qwen3:4b && ollama serve

# Or use LM Studio GUI

2. Run R CLI

# Interactive chat
r

# Direct command
r chat "Explain quantum computing in simple terms"

# Start API server
r serve --port 8765

R OS - Terminal UI (Experimental)

A terminal-based interface that looks like Android. Built with Textual. This is an experimental feature for Raspberry Pi and edge devices - not an actual OS.

┌─────────────────────────────────────────────────────────┐
│ ▁▂▄█ 📶 R OS          12:45          🔋 85%             │
├─────────────────────────────────────────────────────────┤
│                                                         │
│   💬 Messages   📞 Phone     📧 Email     🌐 Browser   │
│                                                         │
│   📷 Camera     🖼️ Gallery   🎵 Music     🎬 Video     │
│                                                         │
│   📁 Files      📅 Calendar  ⏰ Clock     🔢 Calculator │
│                                                         │
│   🤖 R Chat     🎤 Voice     🌍 Translate 📝 Notes     │
│                                                         │
│   ⚙️ Settings   📶 WiFi      🔵 Bluetooth 🔋 Battery   │
│                                                         │
│   💡 GPIO       💻 Terminal  🔌 Network   📊 System    │
│                                                         │
├─────────────────────────────────────────────────────────┤
│           ◀ Back      ● Home      ▢ Recent             │
└─────────────────────────────────────────────────────────┘

Launch

r-os                    # Material theme
r-os --theme amoled     # AMOLED black
r-os --theme light      # Light theme

Keyboard Shortcuts

Raspberry Pi Setup

# One-command installer
curl -sSL https://raw.githubusercontent.com/raym33/r/main/r_os/rpi/install.sh | bash

📖 Full R OS Documentation

All 74 Skills

📄 Documents

pdf · latex · markdown · pdftools · template · resume · changelog

💻 Code & Data

code · sql · json · yaml · csv · regex · schema · diff

🤖 AI & Knowledge

rag · multiagent · translate · faker

🎨 Media

ocr · voice · design · image · video · audio · screenshot · qr · barcode

📁 Files

fs · archive · clipboard · env

📅 Productivity

calendar · email · ical · vcard

🔧 DevOps

git · docker · ssh · http · web · network · system · metrics

🔍 Dev Tools

logs · benchmark · openapi · cron · jwt

📝 Text

text · html · xml · url · ip · encoding

🔢 Data

datetime · color · math · currency · crypto · semver · mime

🌐 Web

rss · sitemap · manifest · hublab · weather

🔌 Hardware (R OS)

gpio · bluetooth · wifi · power · android

🧩 Extensions

plugin

REST API

# Start server
r serve --port 8765

# Chat (OpenAI-compatible)
curl -X POST http://localhost:8765/v1/chat \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "Hello!"}]}'

# Call skill directly
curl -X POST http://localhost:8765/v1/skills/call \
  -d '{"skill": "pdf", "tool": "generate_pdf", "arguments": {"content": "Hello"}}'

Swagger UI: http://localhost:8765/docs

Configuration

# ~/.r-cli/config.yaml
llm:
  backend: ollama
  model: qwen3:4b
  base_url: http://localhost:11434/v1

ui:
  theme: ps2  # ps2, matrix, minimal, retro

skills:
  disabled: []  # Skills to disable

Create Custom Skills

# ~/.r-cli/skills/my_skill.py
from r_cli.core.agent import Skill
from r_cli.core.llm import Tool

class MySkill(Skill):
    name = "my_skill"
    description = "My custom skill"

    def get_tools(self) -> list[Tool]:
        return [
            Tool(
                name="my_function",
                description="Does something useful",
                parameters={"type": "object", "properties": {"input": {"type": "string"}}},
                handler=self.my_function,
            )
        ]

    def my_function(self, input: str) -> str:
        return f"Processed: {input}"

Development

git clone https://github.com/raym33/r.git
cd r
pip install -e ".[dev]"
pytest tests/ -v
ruff check . && ruff format .

Links

• Complete Documentation
• Changelog
• Contributing
• Report Issues
• PyPI Package

Honest Limitations

• Sandboxing is basic - Skills run with your user permissions. Working on better isolation.
• Small models (4B) sometimes pick the wrong tool - Larger models (7B+) work better.
• It's a tool layer, not magic - Prompt quality still matters.
• Some skills need external dependencies - OCR needs Tesseract, voice needs Whisper, etc.

License

MIT License