🛡️ IronClaveBot

The Sovereign, Sandboxed AI Teammate for Developers

An open-source initiative by DevelopersGlobal

Persistent AI Teammates • Zero Host Escapes • Single Static Binary (<10MB) • Embedded Web UI • Zero-Config Local Memory

---

⚡ What is IronClaveBot?

IronClaveBot is an open-source, self-hosted alternative to heavy computer-use bots and proprietary agent workspaces (Grok Bot, Rakazo, OpenClaw).

While existing agent platforms require complicated Node.js runtime versions, heavy PostgreSQL installations, Redis queues, and multi-gigabyte Docker Compose sprawls, IronClaveBot compiles into a single static Go binary with an embedded web dashboard and persistent memory.

🛡️ Why "IronClave"?

Giving an AI agent direct terminal and filesystem access is dangerous. IronClaveBot isolates all tool executions inside a hardened, restricted sandbox enclave:

• Directory Boundary Validation: Agent operations cannot traverse (../) outside the isolated workspace.
• Destructive Command Blocker: Automatic circuit-breakers for harmful system calls (rm -rf /, raw disk writes, fork bombs).
• Zero-Cloud Telemetry: Use local models via Ollama or bring your own API keys. No third-party data tracking.

---

🎥 Demo

1-ironclavebot-live-demo-39f729ed.mp4

Live run: IronClaveBot with a small local model (qwen2.5:0.5b via Ollama) inside a sandboxed container - the agent lists the workspace, reads main.go, and explains the Go project.

---

🚀 10-Second Quickstart

Option 1: Run with Go (Zero Dependencies)

git clone https://github.com/DevelopersGlobal/ironclavebot.git
cd ironclavebot

# Build and start the embedded server
make run

Open http://localhost:8080 in your browser!

Option 2: Run with Docker

docker run -p 8080:8080 -v $(pwd)/workspace:/ironclave/workspace ghcr.io/developersglobal/ironclavebot:latest

Option 3: Terminal CLI Mode (No Browser Required)

./bin/ironclavebot -chat "Inspect workspace files and summarize what you see"

---

📊 How IronClaveBot Compares

Feature	IronClaveBot 🛡️	Rakazo / Grok Bot	AutoGPT / OpenClaw
Setup Time	< 30 seconds	10–20 minutes	15+ minutes
Dependencies	Single Binary (<10MB)	Node 22/24, Postgres, Prisma, Redis	Python, venv, Poetry, Docker
Memory Footprint	~25 MB RAM	~350 MB+ RAM	~500 MB+ RAM
Web Dashboard	Embedded in Binary	Separate React/Vite server	Separate Next.js app
Execution Sandbox	Hardened Enclave Sandbox	Full Docker / VM required	Host process / Unrestricted
License	Apache 2.0 (Open Source)	Apache 2.0	Mixed / Business

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│             Web UI Dashboard (Embedded HTML/JS)             │
└──────────────────────────────┬──────────────────────────────┘
                               │ HTTP / WebSocket REST API
┌──────────────────────────────▼──────────────────────────────┐
│           IronClaveBot Core Engine (Golang Binary)          │
│  ┌─────────────────────────┐   ┌─────────────────────────┐  │
│  │ Persistent Store        │   │ Autonomous ReAct Loop   │  │
│  │ (Embedded SQLite / WAL) │   │ (Multi-Turn Reasoning)  │  │
│  └─────────────────────────┘   └────────────┬────────────┘  │
└─────────────────────────────────────────────┼───────────────┘
                                              │
                      ┌───────────────────────┴───────────────────────┐
                      ▼                                               ▼
         ┌─────────────────────────┐                     ┌─────────────────────────┐
         │   AI Provider Gateway   │                     │  Isolated Sandboxed FS  │
         │ • Ollama (Local/Offline)│                     │ • Path Traversal Guard  │
         │ • OpenAI / DeepSeek     │                     │ • Command Blacklist     │
         │ • Anthropic / Groq      │                     │ • Ephemeral Workspace   │
         └─────────────────────────┘                     └─────────────────────────┘

---

⚙️ Configuration (.env)

Copy .env.example to .env to configure your preferred LLM. .env is read at startup; variables already set in your shell take precedence over it.

# LLM Provider: openai | ollama | mock
LLM_PROVIDER=openai
OPENAI_API_KEY=your_key_here
LLM_MODEL=gpt-4o

# For 100% Offline Local Inference:
# LLM_PROVIDER=ollama
# OLLAMA_BASE_URL=http://localhost:11434
# LLM_MODEL=llama3.2

---

🔐 Security notes

• The dashboard listens on 127.0.0.1 by default. Binding another address (HOST / -host) requires API_TOKEN, and every /api request must then send Authorization: Bearer <token>.
• SECURITY_MODE=interactive (default) and strict refuse command execution and file writes because there is no approval mechanism yet. permissive allows them; use it only inside a container or VM.
• run_command is not an isolation boundary. Commands run as the daemon's user with the host filesystem and network available (only the working directory, a minimal environment, a 30 s timeout, an output cap and a coarse blacklist apply). Run IronClaveBot in a container or VM when giving it a model you do not fully trust.

---

🤝 Community & Contributing

Built with ❤️ by the DevelopersGlobal community.

We welcome all contributions:

• Adding Model Context Protocol (MCP) servers
• Voice streaming (WebRTC / Opus)
• WASM-based plugin extensions
• UI improvements

Read CONTRIBUTING.md to get started!