🤖 Roobie — Autonomous AI Coding Agent

Local-first autonomous AI coding agent · Like Claude Code, but runs offline on 4GB RAM

---

What is Roobie?

Roobie is a terminal-based autonomous AI coding agent powered by AirLLM (disk-offloaded inference).

You type a coding task, and Roobie:

• 🧠 Plans the project
• 📄 Creates & edits files
• 💻 Runs terminal commands
• 🔍 Searches the web
• 🚀 Starts localhost servers
• 🔄 Debugs and improves autonomously

Similar to Claude Code and Gemini CLI, but:

• ✅ Fully offline (no API keys)
• ✅ Runs on 4GB RAM (CPU only)
• ✅ Uses AirLLM (disk-offloaded quantized models)
• ✅ No Ollama required

---

Installation

# 1. Clone and enter the project
cd roobie

# 2. Create virtual environment
python3 -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate

# 3. Install core dependencies
pip install -r requirements.txt

# 4. Install AirLLM + PyTorch (CPU)
pip install airllm transformers accelerate
pip install torch --index-url https://download.pytorch.org/whl/cpu

---

Quick Start

# Start interactive chat (default model: deepseek-coder-1.3b)
python main.py

# Use a specific model
python main.py chat --model Qwen/Qwen2.5-Coder-3B-Instruct

# Set a workspace directory
python main.py chat --workspace ~/my-project

# One-shot task (non-interactive)
python main.py run "Build me a FastAPI REST API with CRUD operations"

# Web search
python main.py search "Next.js 15 new features"

# System status
python main.py status

---

Recommended Models (AirLLM)

Models are auto-downloaded from HuggingFace on first use. AirLLM splits them into chunks and streams from disk — no full RAM loading.

---

CLI Commands

---

Example Session

roobie> Build me a modern SaaS landing page with HTML, CSS, and JavaScript

🧠 Thinking...

📄 Create File → index.html
✓ Created index.html (4.2 KB)

📄 Create File → style.css
✓ Created style.css (2.8 KB)

📄 Create File → app.js
✓ Created app.js (1.1 KB)

💻 Run Command → python3 -m http.server 8080
✓ Server started at http://localhost:8080

🤖 Roobie: Done! I created a modern SaaS landing page with:
- Responsive hero section with gradient
- Features section with cards
- Pricing table
- CTA section
- Smooth scroll animations

Open http://localhost:8080 to see it.

---

Environment Variables

Copy .env.example to .env:

cp .env.example .env

---

Architecture

roobie/
├── main.py               ← Entry point
├── cli/
│   └── app.py            ← Typer CLI (chat loop, slash commands)
├── agent/
│   ├── chat_engine.py    ← Agentic loop (AirLLM + tool calling)
│   └── tool_executor.py  ← Dispatches tool calls
├── tools/
│   ├── file_tools.py     ← Create, edit, read, delete files
│   ├── terminal_tools.py ← Run shell commands safely
│   ├── search_tools.py   ← Web search (no API key)
│   └── thinking_tools.py ← Chain-of-thought reasoning
├── models/
│   └── manager.py        ← AirLLM + Ollama model management
├── browser/
│   └── automation.py     ← Playwright browser automation
├── vision/
│   └── analyzer.py       ← Screenshot analysis (Moondream)
├── memory/
│   └── store.py          ← SQLite conversation memory
└── prompts/
    └── templates.py      ← System prompts

---

Hardware Requirements

Testing

To run the unit test suite, make sure dev dependencies are installed, then run pytest:

# Run all tests
PYTHONPATH=. pytest

# Run settings tests specifically
PYTHONPATH=. pytest tests/test_settings.py

---

License

MIT License — see LICENSE