ClawGo

ClawGo is a Go-native (97.1%) AI agent runtime framework inspired by nanobot (HKUDS), featuring a four-layer architecture (Main Agent → Local Subagents → Runtime Store → WebUI) that provides a complete multi-agent collaboration loop.

Multi-Agent Orchestration#

• Topology: Unified organization of main agent / subagents / remote branches with internal collaboration flow separated from user conversation channel; subagent execution is traceable without polluting the user's main channel
• Default flow: user → main → worker → main → user
• Routing: rules_first mode with keyword-based dispatch to specified subagents, configurable max hops (max_hops: 6) and sticky threads (sticky_thread_owner)
• Communication: mediated mode with message persistence, dead-letter queues, and TTL

Persistence & Resumability#

Full-chain run/event/thread/message persistence via subagent_runs.jsonl, subagent_events.jsonl, threads.jsonl, and agent_messages.jsonl, enabling resumable subagent runs after restart.

Context Management#

Summary-based Context Compaction with configurable trigger message count and recent message retention.

Tool & MCP Integration#

MCP Server support via tools.mcp for stdio, http, streamable_http, and sse transports. Tools are auto-discovered and registered as mcp__<server>__<tool>. Supports permission: workspace or permission: full and parallel tool calls for safe tool lists (tool_max_parallel_calls: 2).

Provider Integration & Authentication#

Supports OpenAI, Codex, Anthropic, Gemini, Kimi, Qwen, and more. Hybrid authentication: API Key priority + OAuth account pool auto-switching (auth: "hybrid").

Observability (WebUI)#

Dashboard with Agent topology view, node/log/memory/runtime state inspection, and OAuth account management. WebUI is read-only and does not write runtime configuration.

Engineering Configuration#

Agent prompts are recommended as separate agents/<name>/AGENT.md files referenced via system_prompt_file. Subagents are declared in config.json → agents.subagents, each with independent session, memory namespace, tool allowlist, and runtime parameters. Routing rules in agents.router.rules, MCP Servers in tools.mcp.servers.

Installation & Deployment#

curl -fsSL https://raw.githubusercontent.com/YspCoder/clawgo/main/install.sh | bash
clawgo onboard
clawgo provider list
clawgo provider use openai/gpt-5.4
clawgo provider configure

Launch modes: clawgo agent (interactive), clawgo gateway run (gateway), make dev (development). WebUI accessible at http://<host>:<port>/?token=<gateway.token>. Dockerfile and docker-compose.yml provided for containerized deployment.

Pending Confirmation#

• Distributed node support mentioned in README title and GitHub description but not detailed in the body
• Official site clawgo.dev currently has limited content; full documentation may still be under construction
• Distinction between Remote Branches and Local Subagents, and Memory storage backend details remain to be confirmed