Quick Start
The Current Recommended Path
According to the current README, the shortest path is:
- install
clawgo - run
clawgo onboard - choose a provider and model
- start
agentmode orgateway run
Install
bash
curl -fsSL https://raw.githubusercontent.com/YspCoder/clawgo/main/install.sh | bashInitialize
bash
clawgo onboardIt will:
- create the default config
- initialize the workspace
- generate
gateway.token
Choose A Provider And Model
The current README path is:
bash
clawgo provider list
clawgo provider use openai/gpt-5.4
clawgo provider configureFor OAuth-backed providers such as codex, anthropic, antigravity, gemini, kimi, and qwen:
bash
clawgo provider login codex
clawgo provider login codex --manualIf the same provider has both an API key and OAuth accounts, auth: "hybrid" is the recommended setup.
Start
Interactive mode:
bash
clawgo agent
clawgo agent -m "Hello"Gateway mode:
bash
clawgo gateway runDevelopment mode:
bash
make devWebUI Access
According to the current README, the WebUI is accessed through Gateway directly:
text
http://<host>:<port>/?token=<gateway.token>That is also the default assumption used in this docs set.
Minimal Config Mental Model
The real config center is now:
agents.defaultsagents.routeragents.communicationagents.subagentsmodels.providers
not the earlier agents.agents wording.
A Minimal Subagent Topology
json
{
"agents": {
"router": {
"enabled": true,
"main_agent_id": "main",
"strategy": "rules_first",
"rules": []
},
"subagents": {
"main": {
"enabled": true,
"type": "router",
"role": "orchestrator",
"system_prompt_file": "agents/main/AGENT.md"
},
"coder": {
"enabled": true,
"type": "worker",
"role": "code",
"system_prompt_file": "agents/coder/AGENT.md"
}
}
}
}First Validation
Start with:
bash
clawgo status
clawgo config checkThen verify that these runtime files begin to appear:
subagent_runs.jsonlsubagent_events.jsonlthreads.jsonlagent_messages.jsonl
A Better First Prompt For The Current Runtime
bash
clawgo agent -m "Implement a new endpoint and let coder and tester split the work"That is closer to the current runtime model:
maindispatches- a
subagentexecutes - internal messages stay separate
- the result comes back to the main session