🪐 Google Antigravity Workspace Template

# 🪐 Google Antigravity Workspace Template

**Production-grade starter kit for autonomous AI agents on Google Antigravity.**

Language: [English](/docs/en/) | [中文（仓库主页）](README_CN.md) | [中文文档](/docs/zh/) | [Español](/docs/es/)

![License](https://img.shields.io/badge/License-MIT-green)
![Gemini](https://img.shields.io/badge/AI-Gemini_2.0_Flash-blue)
![Architecture](https://img.shields.io/badge/Architecture-Event_Driven-purple)
![Memory](https://img.shields.io/badge/Context-Infinite-orange)

## 🌟 Project Intent

In a world full of AI IDEs, I want enterprise-grade architecture to be as simple as **Clone → Rename → Prompt**.

This project leverages IDE context awareness (via `.cursorrules` and `.antigravity/rules.md`) to pre-embed a complete **cognitive architecture** in the repo.

When you open this project, your IDE stops being just an editor—it becomes an **industry-savvy architect**.

**First principles:**

- Minimize repetition: the repo should encode defaults so setup is nearly zero.
- Make intent explicit: capture architecture, context, and workflows in files, not tribal knowledge.
- Treat the IDE as a teammate: contextual rules turn the editor into a proactive architect, not a passive tool.

### Why do we need a thinking scaffold?

While building with Google Antigravity or Cursor, I found a pain point:

**The IDE and models are powerful, but the empty project is too weak.**

Every new project repeats the same boring setup:

- "Should my code live in `src` or `app`?"
- "How do I define utilities so Gemini recognizes them?"
- "How do I help the AI remember prior context?"

This repetition wastes creative energy. My ideal workflow is: **after a git clone, the IDE already knows what to do.**

So I built this project: **Antigravity Workspace Template**.

## ⚡ Quick Start

### Automated Installation (Recommended)

**Linux / macOS:**
```bash
# 1. Clone the template
git clone https://github.com/study8677/antigravity-workspace-template.git my-project
cd my-project

# 2. Run the installer
chmod +x install.sh
./install.sh

# 3. Configure your API keys
nano .env

# 4. Run the agent
source venv/bin/activate
python src/agent.py
```

**Windows:**
```cmd
# 1. Clone the template
git clone https://github.com/study8677/antigravity-workspace-template.git my-project
cd my-project

# 2. Run the installer
install.bat

# 3. Configure your API keys (notepad .env)

# 4. Run the agent
python src/agent.py
```

### Manual Installation

```bash
# 1. Clone the template
git clone https://github.com/study8677/antigravity-workspace-template.git my-project
cd my-project

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

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

# 4. Configure your API keys
cp .env.example .env  # (if available) or create .env manually
nano .env

# 5. Run the agent
python src/agent.py
```

**That's it!** The IDE auto-loads configuration via `.cursorrules` + `.antigravity/rules.md`. You're ready to prompt.

## 🎯 What Is This?

This is **not** another LangChain wrapper. It's a minimal, transparent workspace for building AI agents that:

- 🧠 Have infinite memory (recursive summarization)
- 🛠️ Auto-discover tools from `src/tools/`
- 📚 Auto-inject context from `.context/`
- 🔌 Connect to MCP servers seamlessly
- 🤖 Coordinate multiple specialist agents
- 📦 Save outputs as artifacts (plans, logs, evidence)

**Clone → Rename → Prompt. That's the workflow.**

## 🚀 Key Features

| Feature | Description |
|---------|-------------|
| 🧠 **Infinite Memory** | Recursive summarization compresses context automatically |
| 🛠️ **Universal Tools** | Drop Python functions in `src/tools/` → auto-discovered |
| 📚 **Auto Context** | Add files to `.context/` → auto-injected into prompts |
| 🔌 **MCP Support** | Connect GitHub, databases, filesystems, custom servers |
| 🤖 **Swarm Agents** | Multi-agent orchestration with Router-Worker pattern |
| ⚡ **Gemini Native** | Optimized for Gemini 2.0 Flash |
| 🌐 **LLM Agnostic** | Use OpenAI, Azure, Ollama, or any OpenAI-compatible API |
| 📂 **Artifact-First** | Every task produces plans, logs, and evidence |
| 🔒 **Sandbox Execution** | Configurable code execution environments (local by default) |

## 📚 Documentation

**Full documentation available in `/docs/en/`:**

- **[Quick Start](docs/en/QUICK_START.md)** — Installation & deployment
- **[Philosophy](docs/en/PHILOSOPHY.md)** — Core concepts & architecture
- **[Zero-Config](docs/en/ZERO_CONFIG.md)** — Auto tool & context loading
- **[MCP Integration](docs/en/MCP_INTEGRATION.md)** — External tool connectivity
- **[Swarm Protocol](docs/en/SWARM_PROTOCOL.md)** — Multi-agent coordination
- **[Roadmap](docs/en/ROADMAP.md)** — Future phases & vision

### Sandbox Configuration (Zero-Config by default)

The sandbox lets the agent execute generated Python code safely and consistently. It defaults to a local subprocess with isolation and limits.

- `SANDBOX_TYPE`:

... (truncated)