GoClaw Agent Swarms — OpenClaw Multi-Agent Orchestration

The practical playbook for building GoClaw agent swarms. GoClaw is a multi-tenant AI gateway written in Go — deploy teams of agents wired for Telegram, Discord, Slack, Feishu, Zalo, and WhatsApp, with built-in task boards, inter-agent delegation, skills, heartbeat, cron, and 20+ LLM providers.

---

Table of Contents

• What is GoClaw?
• Why GoClaw?
• Quick Start
• GoClaw Agent Teams
• Agent Pack Templates
• Pack Library
• Porting Packs to GoClaw
• Documentation Map
• Sources

---

What is GoClaw?

GoClaw is OpenClaw rebuilt in Go — a production-grade multi-agent AI gateway with:

• Single binary (~25 MB, no Node.js runtime, <1s startup, runs on a $5 VPS)
• Multi-tenant PostgreSQL — per-user workspaces, encrypted API keys (AES-256-GCM), isolated sessions
• Agent Teams — lead + member orchestration with shared task boards, sync/async delegation, and mailbox messaging
• 20+ LLM providers — Anthropic, OpenAI, OpenRouter, Groq, DeepSeek, Gemini, Mistral, xAI, Ollama, Claude CLI, Codex, and more
• 7 messaging channels — Telegram, Discord, Slack, Feishu/Lark, Zalo OA, Zalo Personal, WhatsApp
• 5-layer security — gateway auth → global tool policy → per-agent → per-channel → owner-only, plus rate limiting and prompt injection detection
• Skills system — SKILL.md packages with BM25 + pgvector hybrid search
• Built-in observability — LLM call tracing with spans, prompt cache metrics, optional OpenTelemetry export
• Heartbeat + Cron — proactive periodic check-ins and scheduled agent tasks
• Knowledge graph — LLM-powered entity extraction and relationship traversal

Full docs at docs.goclaw.sh.

---

Why GoClaw?

	GoClaw	OpenClaw (TypeScript)
Binary size	~25 MB static	28 MB + Node.js
RAM idle	~35 MB	>1 GB
Startup	<1s	>5s
Target hardware	$5 VPS+	$599+ Mac Mini
Multi-tenant	✅ PostgreSQL	✅ File-based
Agent teams	✅ Task board + mailbox	⚠️ Partial
Security	5-layer defense	Basic
20+ LLM providers	✅ Native	10+
Channels	7	37+
Skills system	✅ BM25 + pgvector	✅ Embeddings
Observability	✅ OTLP opt-in	⚠️ Basic

GoClaw gives you production-grade architecture in a tiny footprint.

---

Quick Start

1. Install GoClaw

macOS/Linux:

curl -fsSL https://raw.githubusercontent.com/nextlevelbuilder/goclaw/main/scripts/install-lite.sh | bash

Windows PowerShell:

irm https://raw.githubusercontent.com/nextlevelbuilder/goclaw/main/scripts/install-lite.ps1 | iex

Or from source (Go 1.26+):

git clone -b main https://github.com/nextlevelbuilder/goclaw.git && cd goclaw
make build
./goclaw onboard
source .env.local && ./goclaw

2. Configure at least one LLM provider

Edit .env.local in the goclaw directory:

GOCLAW_ANTHROPIC_API_KEY=sk-ant-...
GOCLAW_OPENAI_API_KEY=sk-...

Or use OpenRouter for access to 100+ models:

GOCLAW_OPENROUTER_API_KEY=sk-or-...

3. Connect a channel

Telegram:

./goclaw channels add telegram
# Follow interactive prompts for bot token

Discord:

./goclaw channels add discord
# Follow interactive prompts for bot token and gateway intents

4. Create your first agent

./goclaw agents add --name "assistant" --provider anthropic

5. Start the gateway

./goclaw
# Web dashboard: http://localhost:18790

---

GoClaw Agent Teams

GoClaw's team system maps directly to the agent pack concept:

Lead Agent (orchestrator)
  ├── Member 1 (specialist)
  ├── Member 2 (specialist)
  ├── Member 3 (specialist)
  ├── Member 4 (specialist)
  └── Member 5 (specialist)

Each agent runs with its own identity, tools, LLM provider, and context files. The lead orchestrates via:

• Task Board (team_tasks) — create, claim, complete tasks with blocked_by dependencies
• Delegation — sync (wait for result) or async (wait for announcement) task assignment
• Mailbox — team members communicate via direct messages

Creating a Team

./goclaw teams create --name "research-desk" --lead research_lead
./goclaw teams add research-desk --agents crawler fact_check data_analyst writer editor

Or via the dashboard at http://localhost:18790.

Delegation Pattern

User → Lead Agent (research_lead)
         ├── Sync delegate → fact_check_lead (quick verification)
         └── Async delegate → crawler_specialist (deep research)
                   ↓ (later)
              crawler_specialist announces result to lead
         ↓
    Lead synthesizes and delivers to user

---

Agent Pack Templates

Pre-built, deployment-ready agent swarms for GoClaw. Each pack is a complete team with role-specific context files, coordination patterns, and Discord/Telegram channel wiring.

Available Packs

Pack	Team	Best For
dev-ops-corp	9 agents	Coding & release workflows
research_report	6 agents	Research → fact-check → polished report
review_desk	6 agents	Multi-specialist review with PASS/FAIL verdicts
financial_analyst	6 agents	Market analysis → risk assessment → briefing
second_brain	7 agents	Personal knowledge capture & retrieval

See templates/samples/ for full pack library, agent rosters, and GoClaw deployment commands.

---

Pack Library

DevOpsCorp — Coding & Release Workflow

templates/samples/dev-ops-corp/

Agent	Role
orchestrator	Planning, intake, task routing
sub1	Execution track A (backend/API)
sub2	Execution track B (automation/scripts)
sub3	Execution track C (docs/ops)
reviewer	QA gate, release readiness
compliance_auditor	DevSecOps & compliance
data_engineer	Data pipeline & ETL
security_architect	Threat modeling, vulnerability review
test_automation	Test strategy & automation

Research Report Pack — 1 Lead + 5

templates/samples/research_report/

Agent	Role
research_lead	Coordinates research, synthesizes findings
crawler_specialist	Web/data harvesting, source identification
fact_check_lead	Claim verification, confidence levels
data_analyst	Statistical processing, pattern detection
technical_writer	Drafts sections, translates for audience
summary_editor	Final polish, formatting, release gate

Review Desk Pack — 1 Lead + 5

templates/samples/review_desk/

Agent	Role
review_lead	Queue management, final verdicts (PASS/FAIL)
technical_reviewer	Code correctness, architecture, edge cases
compliance_reviewer	Regulatory/policy alignment
style_guide_enforcer	Brand voice, formatting, accessibility
quality_assurance	Testing, release readiness sign-off
feedback_curator	Synthesizes and delivers reviewer feedback

Financial Analyst Pack — 1 Lead + 5

templates/samples/financial_analyst/

Agent	Role
finance_lead	Strategic direction, recommendation synthesis
market_researcher	Market conditions, competitive landscape
quantitative_analyst	Statistical modeling, pattern analysis
risk_assessor	Scenario analysis, stress tests, risk matrices
finance_report_writer	Narrative drafting, executive summaries
data_visualizer	Charts, dashboards, presentation visuals

Second Brain Pack — 1 Lead + 6

templates/samples/second_brain/

Agent	Brain Region	Role
orchestrator_agent	Thalamus	Central coordinator, routing
capture_agent	Sensory Cortex	Intake & collection
tagger_agent	Hippocampus	Semantic tagging & categorization
memory_writer_agent	Cortex	Memory consolidation & storage
search_agent	Frontal Lobe	Query processing & recall
recall_agent	Temporal Lobe	Long-term memory retrieval
synthesis_agent	Prefrontal Cortex	Cross-domain synthesis

---

Porting Packs to GoClaw

The 4-file per-agent structure maps directly to GoClaw's context file system:

Original Pack File	GoClaw Context File	Purpose
SOUL.md	SOUL.md	Persona, core truths, tone, boundaries
IDENTITY.md	IDENTITY.md	Name, emoji, avatar, communication style
AGENTS.md	(integrated into SOUL.md + system prompt)	Workspace conventions, memory, group chat rules, heartbeat
TOOLS.md	TOOLS.md	Role-specific tool notes, environment details
USER.md	USER.md	Human profile, project context, preferences

Key Differences

1. No file-based workspaces — GoClaw stores agents, context files, and memory in PostgreSQL. Files are loaded into context at runtime.

2. Teams replace orchestration scripts — Instead of Discord channel routing and manual agent coordination, use GoClaw's native team system with team_tasks and delegation.

3. Skills are first-class — Pack-specific skills (e.g., SKILL.md for research methodology, financial analysis, code review) go in goclaw/skills/ and are injectable per-agent.

4. TEAM.md is auto-generated — GoClaw injects TEAM.md at runtime for team members. Don't create it manually.

5. Context file injection — Use ./goclaw agents update <name> --context-files or the dashboard to inject pack context files into agents.

Porting Steps

1. Map agents to GoClaw team members
   agents add --name research_lead --provider anthropic
   agents add --name crawler_specialist --provider openai
   ...

2. Create a team and link agents
   teams create --name research-report --lead research_lead
   teams add research-report --agents crawler_specialist fact_check_lead data_analyst technical_writer summary_editor

3. Inject context files per agent
   agents update research_lead --context-files templates/samples/research_report/research_lead/

4. Add pack-specific skills
   cp -r templates/samples/research_report/skills/ goclaw/skills/

5. Configure channel bindings
   agents bind --agent research_lead --channel telegram --chat-id YOUR_CHAT_ID

See GOCLAW_PACKS.md for the full porting guide.

---

Documentation Map

Start Here

• GOCLAW_SETUP.md — Complete installation and configuration guide
• GOCLAW_PACKS.md — Porting agent packs to GoClaw teams
• Full docs: docs.goclaw.sh

Core Reference

• Teams: docs.goclaw.sh/#teams-what-are-teams
• Agents: docs.goclaw.sh/#agents-explained
• Skills: docs.goclaw.sh/#skills
• Channels: docs.goclaw.sh/#channels-overview
• Providers: docs.goclaw.sh/#providers-overview

Templates (GoClaw-Native)

• templates/samples/ — Agent pack templates with GoClaw-compatible context files
• templates/AGENTS.md — Default AGENTS.md (GoClaw-compatible)
• templates/SOUL.md — SOUL.md template
• templates/IDENTITY.md — IDENTITY.md template
• templates/USER.md — USER.md template

Legacy (OpenClaw/AgentSwarm)

• agentsswarm/ — Original TypeScript/React AgentSwarm runtime (deprecated)
• CLAW_EMPIRE_SETUP.md — Legacy visual office reference

---

Sources

• GoClaw GitHub: github.com/nextlevelbuilder/goclaw
• GoClaw Docs: docs.goclaw.sh
• GoClaw Lite (Desktop): docs.goclaw.sh/#desktop-edition-goclaw-lite
• Agent Teams: docs.goclaw.sh/#teams-what-are-teams
• Skills: docs.goclaw.sh/#skills