🚀 Antigravity-Code v2.2.0

Multi-API Agentic AI Coding Assistant

Agentic Workflow • RAG Memory • Vision • Local Models • Smart Failover

Never lose productivity to API downtime again!

Features • Installation • Quick Start • New in v2.0 • Documentation

🌟 Overview

Antigravity-Code is a powerful terminal-based AI coding assistant designed for Windows developers. Unlike traditional AI assistants that rely on a single provider, Antigravity features intelligent multi-API architecture with automatic failover, ensuring uninterrupted coding assistance even when individual AI providers experience downtime.

With v2.0, Antigravity evolves into a fully Agentic System. It doesn't just write code; it plans, implements, reviews, and fixes it for you using a team of specialized AI agents. It also gains sight (Vision), memory (RAG), and complete privacy with Local Models (Ollama).

Why Antigravity?

🔄 99.9% Uptime - Automatic failover across 3 AI providers
🤖 Agentic Workflow - Autonomous Planner, Coder, and Reviewer agents
🧠 RAG Memory - Indexes your codebase for intelligent context retrieval
👁️ Vision & UI - Converts screenshots directly to HTML/CSS code
🏠 Local & Private - Run 100% offline with Ollama (Llama 3, Mistral, etc.)
🎯 85+ AI Models - Choose from Gemini, Claude, and OpenAI
⏮️ Checkpoint System - Time-travel through file changes with snapshots
🔒 Permission Modes - 3 control modes (default, auto-edit, plan-only)
📦 Batch Operations - Visual file tree with multi-file approval
🌿 Session Forking - Branch conversations without losing context
🛡️ Zero Context Loss - Seamless provider switching mid-conversation
🔒 Privacy First - Local storage, encrypted keys, no telemetry
⚡ Lightning Fast - Native Windows terminal integration
🎨 Beautiful CLI - Colorized output with progress indicators

✨ Features

Core Capabilities

Feature	Description
🤖 Agentic Mode	Autonomous planning, coding, and reviewing workflow
💻 Code Generation	Write clean, efficient code in any language
👁️ Vision Analysis	Analyze images and generated code from screenshots
🧠 RAG Context	Index entire codebase for accurate answers
📂 File System Ops	Read, write, and delete files directly
⏮️ Checkpoint System	NEW! Time-travel file snapshots with `/rewind` and `/checkpoints`
🔒 Permission Modes	NEW! 3 control modes: default, auto-edit, plan-only
📦 Batch Operations	NEW! Visual file tree with multi-file approval/rejection
🌿 Session Forking	NEW! Branch conversations with `/fork` command
📊 Context Management	NEW! Smart compaction with `/compact` and `/context`
🐛 Intelligent Debugging	Find and fix bugs systematically using Linter and Auto-Fix
🔄 Smart Commit	Auto-generate conventional commit messages from git diff
🧪 Test Generation	Automated unit tests written to disk
📝 Documentation	Auto-generate comprehensive docs
🔄 Refactoring	Improve code quality and structure
🧠 Context Awareness	Auto-detects project type (Node, Python, etc.)
🏗️ Project Scaffolding	Auto-generate Fractal Agent structure with `/init`
🛡️ Self-Healing	Autonomous error detection and fixing loop
📦 Safe Execution	Sandboxed execution with blocklist & user confirmation
🔌 IDE Integration	MCP Server & JSON output for VS Code/Cursor

Advanced Features

🏠 Local Model Support: Full support for Ollama (Llama 3, etc.) for offline coding.
🔄 Intelligent Failover: Automatic switching between Gemini → Claude → OpenAI.
🎨 Rich CLI: Vibrant colors, animated spinners, and markdown rendering.
🐚 Shell Integration: Native ag alias and system PATH setup.
📊 Context Management: Smart conversation history with 5MB compression.
🛠️ Extensible Skills: Custom skills for specialized tasks.
🚀 Workflow Automation: Pre-defined workflows (/create, /debug, /test).
🔐 Secure Storage: Windows DPAPI encryption for API keys.

New in v2.2.0 🎉

⏮️ Checkpoint System: Automatic file snapshots before every edit. Rewind to any previous state with /rewind <checkpoint_id>. List all checkpoints with /checkpoints.
🔒 Permission Modes: Control AI behavior with 3 modes:
- default: Ask before each action
- auto-edit: Automatically apply all changes
- plan-only: Show plans without executing
📦 Visual File Tree: Beautiful tree visualization for multi-file operations with status icons (✏️ Modified, ➕ New, ❌ Deleted).
🎯 Batch Operations: When AI proposes multiple file changes, see them all at once and choose:
- [A] Apply All - Accept all changes instantly
- [R] Review Each - Review files one by one
- [C] Cancel - Reject all changes
🌿 Session Forking: Branch your conversation with /fork to explore different approaches without losing the original context.
📊 Context Management: Smart context compaction prioritizes important messages. Use /compact to optimize and /context to view statistics.
🛡️ Security Audit: Scan your code for vulnerabilities and quality issues with /audit.
🕰️ Session Management: Save, load, and manage your conversation history with /session.
🏗️ Smart Init: Generate project scaffolding with AI-driven content using /init.

📦 Installation

Prerequisites

Node.js 18.0.0 or higher (Download)
npm 9.0.0 or higher
Windows 10/11 with PowerShell 5.1+
Git installed and available in PATH
(Optional) Ollama for local models (Download)
API Key from at least one provider (if not using Ollama):
- Google AI Studio (Gemini - Recommended)
- Anthropic Console (Claude)
- OpenAI Platform (OpenAI)

Quick Install

# Clone the repository
git clone https://github.com/gkhantyln/antigravity-code.git
cd antigravity-code

# Install dependencies
npm install

# Run setup wizard
npm run setup

# Integrate with shell (Path & Aliases)
npm run integrate

The setup wizard will guide you through:

✅ API key configuration (encrypted storage)
✅ Model selection for each provider
✅ Environment setup
✅ Shell integration (ag alias)

🚀 Quick Start

Interactive Mode (REPL)

ag
# or
antigravity-code

🐳 Docker Quick Start

Run Antigravity in a container without installing Node.js:

# Build and run
docker-compose up -d
docker-compose exec antigravity ag

Example Session:

🚀 Antigravity-Code v2.0
Connected to: gemini (gemini-2.5-flash)

AG> /agent "Refactor the auth middleware to use JWT"

🤖 Planner: Creating execution plan...
1. [ ] Install jsonwebtoken package
2. [ ] Update middleware/auth.js
3. [ ] Update routes/auth.js

🤖 Coder: Executing step 1...
✓ Installed jsonwebtoken

🤖 Reviewer: Checking code changes...
✓ Linting passed. functionality verified.

> /commit
AI: Suggested message: "refactor(auth): migrate to JWT authentication"
Commit? (y/n): y
✓ Committed.

Single Command Mode

ag "Explain async/await in JavaScript"

Pipe Mode

git diff | ag "Explain these changes"
cat error.log | ag "Debug this error"

🎯 Available Commands

Interactive Commands

Command	Description	Example
`/agent`	Start autonomous agentic workflow	`/agent "Build a todo app"`
`/ui`	Convert screenshot to HTML/CSS code	`/ui mockup.png`
`/see`	Analyze an image	`/see error_screenshot.png`
`/commit`	Smart git commit message generation	`/commit` (auto-detects changes)
`/index`	Index codebase for RAG	`/index`
`/init`	Initialize Fractal Agent scaffolding	`/init my-app`
`/rewind`	NEW! Revert file to checkpoint	`/rewind abc123`
`/checkpoints`	NEW! List recent file checkpoints	`/checkpoints`
`/permission`	NEW! Change permission mode	`/permission auto-edit`
`/session`	NEW! Manage conversation history	`/session save "My Feature"`
`/audit`	NEW! Audit code for security/quality	`/audit src/app.js`
`/fork`	NEW! Fork current conversation	`/fork`
`/compact`	NEW! Compact conversation context	`/compact`
`/context`	NEW! Show context statistics	`/context`
`/create`	Create new features/files directly	`/create "snake game in python"`
`/debug`	Debug and fix code issues	`/debug "fix this error"`
`/test`	Generate and save unit tests	`/test "app.js"`
`/model`	Change Gemini/Ollama model	`/model llama3`
`/provider`	Switch AI provider	`/provider ollama`
`/config`	View/update configuration	`/config`
`/new`	Start new conversation	`/new`
`/clear`	Clear screen	`/clear`
`/help`	Show help	`/help`
`/exit`	Exit Antigravity	`/exit`

🤖 Supported Models

Gemini (Primary Provider) - 14 Models

Gemini 3 Series (Latest)

gemini-3-flash - Fast & fluid multimodal
gemini-3-pro - Most powerful general model
gemini-3-pro-image - Visual + text
gemini-3-deep-think - Advanced analysis

Gemini 2.5 Series

gemini-2.5-pro - Production & powerful
gemini-2.5-flash ⭐ Default - Fast & economical
gemini-2.5-flash-tts - Text-to-speech

Legacy Models

gemini-2.0-flash-exp, gemini-2.0-flash, gemini-2.0-flash-lite
gemini-1.5-pro, gemini-1.5-flash, gemini-1.0-pro

Ollama (Local Provider) - NEW!

llama3
mistral
codellama
Any model pulled via ollama pull <model>

Claude (Secondary Provider) - 11 Models

Opus Series (Most Powerful)

claude-opus-4.6 - Latest & most capable
claude-opus-4.1, claude-opus-4

Sonnet Series (Balanced)

claude-sonnet-4.5 - Recommended
claude-sonnet-4, claude-sonnet-3.7, claude-sonnet-3.5

Haiku Series (Fast & Economical)

claude-haiku-4.5, claude-haiku-3.5, claude-haiku-3

OpenAI (Tertiary Provider) - 60+ Models

GPT-5.x Series

gpt-5.3-codex, gpt-5.3-codex-spark (Latest)
gpt-5.2, gpt-5.2-pro, gpt-5.2-codex
gpt-5.1, gpt-5.1-codex, gpt-5.1-codex-max
gpt-5, gpt-5-pro, gpt-5-codex, gpt-5-mini, gpt-5-nano

o-Series (Reasoning)

o3, o3-pro, o3-mini, o4-mini-deep-research

GPT-4.1 Series

gpt-4.1, gpt-4.1-mini, gpt-4.1-nano

Specialized Models

Multimodal: gpt-image-1, gpt-image-1-mini
Audio: gpt-audio, gpt-realtime, whisper
Embedding: text-embedding-3-large, text-embedding-3-small
Utility: tts-1, tts-1-hd, omni-moderation

⚙️ Configuration

Environment Variables

Edit .env file:

# Provider Priority
PRIMARY_PROVIDER=gemini # or ollama
SECONDARY_PROVIDER=claude
TERTIARY_PROVIDER=openai

# Gemini Configuration
GEMINI_DEFAULT_MODEL=gemini-2.5-flash

# Ollama Configuration (NEW)
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama3

# Failover Settings
FAILOVER_ENABLED=true
MAX_RETRIES_PER_PROVIDER=3
RETRY_DELAY_MS=1000
HEALTH_CHECK_INTERVAL_MS=30000

# Context Settings
MAX_CONVERSATION_MESSAGES=50
MAX_FILE_CONTEXT=10
MAX_CONTEXT_SIZE_MB=5

Model Switching

Switch to Local Mode:

> /provider ollama
✓ Switched to provider: ollama

> /model llama3
✓ Changed Ollama model to: llama3

🛡️ Security

Antigravity-Code prioritizes your security and privacy:

🔐 Encrypted API Keys - Windows DPAPI encryption
🏠 Local Mode - Use Ollama for complete data privacy (no data leaves your machine)
💾 Local Storage - All data stored locally on your machine
🚫 No Telemetry - Zero data collection or analytics
🔒 Secure Logging - Automatic secret redaction in logs
🛡️ Input Sanitization - Protection against injection attacks

API Key Storage:

Location: C:\Users\<username>\.antigravity\keys.json
Encryption: Windows DPAPI (tied to your user account)
Security: Keys cannot be transferred to other machines/users

📚 Documentation

User Guides

Quick Start Guide - Get started in 5 minutes
Configuration Guide - Advanced settings
Workflow Guide - Using workflows effectively

Developer Documentation

Architecture Overview - System design
API Reference - API documentation
Contributing Guide - How to contribute

🎨 Examples

Example 1: Create a REST API

> /create
AI: What feature would you like to create?
You: REST API for user management with JWT auth
AI: I'll create `src/routes/auth.js` and `src/controllers/authController.js` with:
- User CRUD operations
- JWT authentication
- Input validation
Proceed? (y/n): y
✓ Files created successfully.

Example 2: Debug an Error

> /debug
AI: What issue are you experiencing?
You: Getting "Cannot read property 'map' of undefined"
AI: Let me analyze...
[Provides detailed analysis and fix]

Example 3: Generate Tests

> /test "src/utils/validator.js"
AI: I will generate unit tests for `validator.js`.
✓ Created `tests/validator.test.js` with 5 test cases.

Example 4: Agentic Workflow (NEW)

> /agent "Implement a User Profile page with React"
🤖 Planner: Creating execution plan...
1. [ ] Create Profile.js component
2. [ ] Add CSS definitions
3. [ ] Update App.js router
...

Example 5: Vision Analysis (NEW)

> /see diagram.png "Explain this database schema"
AI: This diagram shows a relational database with 3 tables...

🏗️ Project Structure

antigravity-code/
├── .agent/                    # Fractal Architecture
│   ├── GEMINI.md             # Agent configuration
│   ├── rules/                # Coding rules
│   ├── skills/               # AI skills
│   └── workflows/            # Automation workflows
├── src/
│   ├── api/                  # API providers
│   │   ├── gemini.js        # Gemini (14 models)
│   │   ├── claude.js        # Claude (11 models)
│   │   ├── openai.js        # OpenAI (60+ models)
│   │   ├── ollama.js        # Ollama (Local Models)
│   │   └── orchestrator.js  # Failover logic
│   ├── cli/                  # CLI interface
│   ├── core/                 # Core engine
│   ├── tools/                # AI Tools
│   │   ├── filesystem.js    # File System ops
│   │   ├── git.js           # Git operations
│   │   └── linter.js        # Linter integration
│   └── utils/                # Utilities
├── scripts/
│   └── setup.js              # Setup wizard
├── test/                     # Tests
└── package.json

📊 Roadmap

Version 2.2.0 (Current) ✅

✅ Checkpoint System (File snapshots with /rewind and /checkpoints)
✅ Permission Modes (3 modes: default, auto-edit, plan-only)
✅ Session Forking (Branch conversations with /fork)
✅ Context Compaction (Smart memory management)
✅ Visual File Tree (Beautiful tree rendering with status icons)
✅ Batch Operations (Multi-file approval with [Apply All] [Review Each] [Cancel])
✅ Session Management (List, Load, Save, New with /session)
✅ Security Auditing (Code analysis with /audit)
✅ Smart Scaffolding (AI-driven /init with context)

Version 2.1.0 ✅

✅ Autonomous Self-Healing (Auto-fix errors)
✅ Safe Execution (Sandboxing & Blocklist)
✅ UX Polish (Colorized Diffs & Confirmations)
✅ IDE Readiness (MCP Server & JSON Output)

Version 2.0 (Current) ✅

✅ Agentic Workflow (Planner, Coder, Reviewer)
✅ RAG Memory (Codebase Indexing)
✅ Vision Support (Image Analysis & UI Gen)
✅ Local Models (Ollama Integration)
✅ Git Integration (Smart Commit)
✅ Proactive Debugging (Linting)
✅ Project Scaffolding (Fractal Agent /init)

Version 1.1 (Completed) ✅

✅ Rich CLI with chalk & ora
✅ Deep Shell Integration (ag alias)
✅ Intelligent Context Awareness
✅ Markdown Rendering

Version 1.0 (Completed) ✅

✅ Multi-API support (Gemini, Claude, OpenAI)
✅ 85+ AI models
✅ Intelligent failover
✅ Terminal interface
✅ Skills and workflows

🤝 Contributing

We welcome contributions! Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests.

We use Changesets for version management and Husky for git hooks to ensure code quality.

🐛 Troubleshooting

Common Issues

Problem: "Ollama connection failed"

# Solution: Ensure Ollama is running
ollama serve

Problem: "Invalid API key"

# Solution: Re-run setup
npm run setup

For more help, see Troubleshooting Guide

📄 License

🙏 Acknowledgments

Special thanks to:

Google AI for Gemini API
Anthropic for Claude API
OpenAI for OpenAI API
Ollama for Local LLM support
All contributors and supporters of this project

🥊 The Ultimate Showdown: Antigravity vs Claude Code

Why limit yourself to one brain when you can have the entire hive mind? Antigravity-Code is engineered to be Anti-Fragile, correcting itself and adapting to any situation.

Feature Strategy	🚀 Antigravity-Code (The Hive Mind)	🤖 Claude Code (The Solitary Cloud)
🧠 Intelligence Architecture	Fractal Agent Swarm (Planner + Coder + Reviewer agents working in parallel)	Linear Chain (One-step thinking)
🛡️ Resilience Strategy	Unstoppable Failover (Auto-switches: Gemini → Claude → OpenAI if one fails)	Single Point of Failure (If Anthropic is down, you stop coding)
🔒 Data Sovereignty	100% Private & Offline Capable (Run locally via Ollama with zero data leak)	Cloud Locked (Your code always leaves your machine)
⚡ Performance	Native Shell Speed (Interacts directly with OS kernel & filesystem)	API Latency Dependent
👁️ Sensory Input	True Multi-Modal Vision (Sees screenshots, diagrams, and UI mockups)	Text-Based constraints
🔄 DevOps Integration	Smart Context Awareness (Auto-detects environment, git state, and project type)	Standard CLI inputs
💸 Economy	Cost-Agnostic (Switch to free local models to save $)	Pay-per-token only
🏗️ Scaffolding	Fractal Core Injection (`/init`) (Instantly builds enterprise-grade agent structure)	Basic file creation

🏆 The Verdict: Antigravity-Code Dominates

While Claude Code is a brilliant tool, Antigravity-Code is a Platform. It gives you the freedom of Open Source, the power of Multi-Model orchestration, and the security of Local Execution. Don't just code; Antigravity.

📞 Contact & Support

Get in Touch

Need help?

📧 Email: tylngkhn@gmail.com
🐛 Issues: GitHub Issues
💬 Discussions: GitHub Discussions
💼 LinkedIn: Gökhan TAYLAN

Developed with 💡 by gkhantyln

Made with ❤️ for developers who never stop coding

⬆ Back to Top

Name		Name	Last commit message	Last commit date
Latest commit History 41 Commits
.changeset		.changeset
.github		.github
.husky		.husky
bin		bin
docs		docs
scripts		scripts
src		src
test		test
.env.example		.env.example
.eslintrc.json		.eslintrc.json
.gitignore		.gitignore
.mocharc.json		.mocharc.json
.prettierrc		.prettierrc
.prettierrc.json		.prettierrc.json
CONTRIBUTING.md		CONTRIBUTING.md
Dockerfile		Dockerfile
README.md		README.md
docker-compose.yml		docker-compose.yml
package-lock.json		package-lock.json
package.json		package.json

gkhantyln/antigravity-code

Folders and files

Latest commit

History

Repository files navigation