Concept Overview

This section answers a simple question: what exactly ClawGo is, and why it is designed this way.

If you think of ClawGo as only a chatbot, you will misread its boundaries. From the current codebase, it is much closer to a long-running agent runtime with three main goals:

• Make agents execute continuously instead of only replying
• Make multi-agent collaboration visible, traceable, and recoverable
• Make prompts, tools, roles, channels, and scheduled tasks manageable as engineering assets

Suggested Reading Order

1. Architecture
2. Runtime, Storage, and Recovery
3. Subagents and Skills

What To Understand Here

1. Runtime, Not Chat Shell

ClawGo has its own:

• AgentLoop
• message bus
• sessions and threads
• persistence and recovery
• hot reload and operations surface

That is why it is not just “user says one thing, model says one thing back.”

2. Multi-Agent Collaboration Is First-Class

The configuration, WebUI, and runtime all treat these as core entities:

• main agent
• subagent
• node-backed branch
• thread
• message
• task

3. Recovery and Observability Are Default Capabilities

From jsonl audit logs and sessions to logs, task audit, and EKG, the system is clearly shaped for production-oriented runtime behavior, not just demos.