The Go-Powered AI Agent Harness — Built for Real Execution, Not Just Demos.
⭐ The Go implementation of next-generation AI agent systems
⭐ Inspired by the architecture behind modern AI coding agents
⭐ Built for performance, extensibility, and real-world automation
In 2026, the internal architecture of modern AI coding agents became publicly understood.
What emerged was not just a tool —
but a new paradigm:
👉 AI is no longer just generating text — it is executing systems.
Projects like claw-code demonstrated:
- Tool orchestration
- Multi-agent execution
- Runtime context management
- Plugin-based architectures
These systems represent the operating system of AI agents. :contentReference[oaicite:0]{index=0}
Claw-Go is a clean-room Go implementation of an AI Agent Harness.
It is NOT:
- ❌ a wrapper around an LLM
- ❌ a prompt engineering demo
- ❌ a toy chatbot
It IS:
✅ A runtime system for AI agents
✅ A tool execution engine
✅ A foundation for autonomous systems
LLM is NOT the system. LLM is just the brain.
The system = Runtime + Tools + Memory + Orchestration
🏗 Architecture Overview
┌──────────────────────┐
│ CLI │
└─────────┬────────────┘
↓
┌──────────────────────┐
│ Commands │
└─────────┬────────────┘
↓
┌──────────────────────┐
│ Query Layer │
└─────────┬────────────┘
↓
┌────────────────────────────────────┐
│ Runtime | Plugins | Tools | MCP │
└────────────────────────────────────┘
↓
┌──────────────────────┐
│ LLM Providers │
└──────────────────────┘
claw-go/
├── cmd/claw # CLI entry
├── internal/
│ ├── app # Command dispatcher
│ ├── models # Shared data structures
│ ├── manifest # System manifest
│ ├── query # Query + output layer
│ ├── commands # Command registry
│ ├── tools # Tool system
│ ├── runtime # Agent runtime core
│ ├── plugins # Hook pipeline
│ ├── providers # LLM abstraction
│ └── mcp # MCP registry
└── go.mod
git clone https://github.com/shirdonl/claw-code-go.git
cd claw-code-go
go run ./cmd/claw summary
go run ./cmd/claw manifest
go run ./cmd/claw subsystems --limit 16
go run ./cmd/claw commands --limit 10
go run ./cmd/claw tools --limit 10
go run ./cmd/claw parity-audit
🧠 Runtime Session lifecycle Context management Execution loop
Plugin-based architecture Dynamic tool invocation Extensible capabilities
pipeline := plugins.New()
pipeline.Use(func(input string) string {
return "[hook]" + input
})
Hook pipeline Middleware injection Event processing
type Provider interface {
Generate(prompt string) string
}
Multi-LLM support OpenAI / DeepSeek / Local models Streaming-ready
External tool integration Registry-based architecture Future: distributed agent systems
Most agent frameworks are:
Python → slow, hard to scale TypeScript → complex runtime
Claw-Go is built for:
⚡ High performance 🧵 Concurrency (goroutines) 🔒 Memory safety 🚀 Production systems 🧪 Current Status
| Feature | Status |
|---|---|
| CLI | ✅ Ready |
| Commands | ✅ Ready |
| Tool System | 🟡 Basic |
| Runtime | 🟡 Skeleton |
| Plugins | 🟡 Basic |
| Providers | 🟡 Mock |
| MCP | 🟡 Skeleton |
Full Agent Runtime Loop Tool Auto Invocation Multi-Agent System DeepSeek / OpenAI Integration Streaming Execution Web UI
claw run "Analyze this repository and generate documentation"
We believe:
The future of software is AI-native systems
Claw-Go aims to become:
The Go LangGraph The Go AutoGen The Agent OS for Developers
This project:
Does NOT contain proprietary code Is a clean-room implementation Is not affiliated with any original proprietary system ⭐ Star History
If this project helps you:
👉 Give it a ⭐ 👉 Share it 👉 Build something crazy with it
🤝 Contributing
PRs welcome.
We are building:
The next generation of AI systems.
LLMs are not the revolution. Systems built around them are.