Note
Disclaimer: This project, "Antigravity (AG)", is an independent template for Agentic AI development and is not affiliated with, endorsed by, or connected to Google's "Antigravity" project or deep research initiatives. "Antigravity" here refers to the concept of lifting the burden of maintenance through AI.
🚀 Antigravity: The definitive template for building production-grade Autonomous Agents. Hybrid architecture, dual-layer memory (ChromaDB + Filesystem), sequential thinking, and battle-tested guardrails. From concept to deployment in minutes.
Advanced project template designed for building robust, replicable, and self-correcting AI Agents. Bridges high-level LLM reasoning with deterministic Python/Node execution. Features dual persistence memory, strict security protocols, and a complete toolkit of workflows, skills, and evaluators.
Mission: Provide the "Gold Standard" operational foundation for AI Engineering, eliminating the fragility of traditional agent "demos".
| Feature | Description |
|---|---|
| Hybrid Architecture | Seamlessly integrates Python (Backend/Tools), Node.js (Frontend), and Agentic Workflows |
| Dual Persistence Memory | Combines ChromaDB (Vector Search) with Filesystem Logs (Human-readable JSON/Markdown) |
| Sequential Thinking | Enforces "System 2" reasoning (Think → Plan → Act) before any critical operation |
| Strict Guardrails | Pre-configured rules for Security, Testing (Regression Gate), and Quality |
| Hydration Ready | Clone and customize for any project via interactive wizard in minutes |
| Resume Capability | Hot State persistence (STATE.md) survives crashes and context resets |
| English-First | All rules, workflows, and logs in English for optimal AI comprehension |
.
├── .agent/ # The "Brain": Memory, Rules, Skills, Workflows
│ ├── memory/ # Hot State (STATE.md) & Cold Logs
│ ├── audit/ # Machine Logs & Human Summaries (JSONL + MD)
│ ├── fix_logs/ # Persistent technical fix records (JSON)
│ ├── evals/ # Capability evaluation suite (Triad: Refactor/Bugfix/Feature)
│ ├── rules/ # Active constraints & protocols
│ ├── skills/ # Executable toolkits (security_audit, fixlog_writer, etc.)
│ ├── tools/ # Python utilities (canary_check, librarian, etc.)
│ └── workflows/ # Standard Operating Procedures (.md)
├── docs_custom/ # 🧠 Project Intelligence (The "Soul")
│ ├── SOURCES.md # Canon Source hierarchy & Gold references
│ ├── domain_language.md # Ubiquitous Language & definitions
│ ├── product_strategy.md # Vision, OKRs & Roadmap
│ ├── brand_identity_guide.md # UX/UI Guidelines & tokens
│ └── architecture.md # Technical design & ADRs
├── LICENSE # MIT License
└── README.md # This file- OS: Ubuntu 22.04 / 24.04 LTS (Recommended)
- Python: Managed via
uv(Fast Python package manager) - Node.js: LTS (Optional, for frontend projects)
git clone https://github.com/Mic23-01/antigravity-template.git my-project
cd my-projectuv run .agent/tools/canary_check.py --no-sentinelTip
Custom Brain Path: Set AG_BRAIN_DIR environment variable to override the default brain location (~/.gemini/antigravity/brain). This is useful for custom deployments or CI environments.
Invoke your AI agent and run:
@[/custom_project]
Follow the interactive wizard to generate your docs_custom/ and project-specific configuration.
| # | Protocol | Description |
|---|---|---|
| 1 | No "Raw" Thinking | Every complex task starts with sequential-thinking |
| 2 | English-First | All Rules, Workflows, and System Logs in English |
| 3 | Resume Capability | Agent maintains "Hot State" (STATE.md) to survive crashes |
| 4 | Verify First | Use RAG tools (brave_search, markdownify) to validate before coding |
| 5 | Fail Fast | Operations halt immediately if Regression Gate or Security Audit fails |
| 6 | Dual Persistence | Every fix logged to both ChromaDB and filesystem for recoverability |
| Workflow | Purpose |
|---|---|
/tech_rag |
Technical tasks with adaptive complexity (Strict/Adaptive modes) |
/research_rag |
Deep research with structural impact analysis (DuckDB) |
/refactor |
Large-scale refactoring with AST analysis and blast radius |
/custom_project |
Interactive hydration wizard for new projects |
/librarian |
Proactive hygiene, ghost code detection, documentation sync |
| Skill | Function |
|---|---|
regression_gate |
Unified validator for Chroma & Filesystem integrity |
fixlog_writer |
Standardized JSON log generation with ChromaDB persistence |
security_audit |
Secret scanning, risky file detection, dependency checks |
test_gate_bivio |
Interactive test depth selector (Smoke/Deep/Debug) |
resolve_canon_sources |
Source hierarchy resolution (Custom > Template) |
ui_ux_designer |
Design database access for colors, typography, styles |
excalidraw_canvas |
🎨 Visual reasoning: live diagrams via Excalidraw (v3.0.0+) |
- ✅ Zero Silence Policy: Pre-existing errors reported immediately
- ✅ Atomic Changes: Small, localized, verifiable modifications
- ✅ Security Guardrails: No
.envleakage, secrets scanning, scope limits - ✅ Evidence Bundle: Every task outputs what changed, tests run, and FixLog ID
Note
This is an optional visual reasoning capability requiring external setup.
The excalidraw_canvas skill enables AI agents to create live architectural diagrams.
Quick Start (Docker):
# Canvas Server
docker run -d -p 3000:3000 --name mcp-canvas ghcr.io/yctimlin/mcp_excalidraw-canvas:latest
# Set environment variable
export EXCALIDRAW_MCP_PATH="$HOME/mcp_excalidraw/dist/index.js"For complete setup options, see: excalidraw_canvas/SKILL.md
- Python 3.10+
uv(recommended) orpip- ChromaDB (optional, for vector memory)
- Node.js 18+ (optional, for Excalidraw)
- GitHub Issues: Bug reports and feature requests
- Workflows Documentation: Complete guides in
.agent/workflows/ - Contributions: Welcome! Follow the English-First policy for PRs
MIT License - see LICENSE file for details.
Project Founder: MIC | Version: 3.0.1 | Status: Production Ready