Last updated: March 2026
Getting Started
Claw-Stack is not a single package β it is a collection of modules built on top of OpenClaw. The first step is always getting OpenClaw running. Everything else layers on after that.
Private beta
Claw-Stack is an active personal research project. The modules described here are open source but not packaged as a turnkey product. For questions or collaboration, reach out at [email protected].
Prerequisites
| Requirement | Notes |
|---|---|
| Node.js 22+ | Required by OpenClaw |
| npm | For installing OpenClaw globally |
| An API key | Anthropic, OpenAI, Gemini, or another OpenClaw-supported provider |
| macOS or Linux | Windows is not officially supported by OpenClaw |
Step 1 β Install OpenClaw
Install OpenClaw globally via npm. OpenClaw is the Node.js gateway that Claw-Stack modules build on top of.
npm install -g openclaw
openclaw --version
For detailed installation instructions, pairing flows, and platform-specific notes, refer to the official OpenClaw documentation.
Step 2 β Start the Gateway
The OpenClaw gateway is the central process that manages agents, channels, and tool execution. Start it with:
openclaw gateway start
On first start, OpenClaw creates ~/.openclaw/ and populates it with a default workspace. The gateway config lives at ~/.openclaw/openclaw.json.
You can also run openclaw onboard for a guided wizard that walks through pairing your first channel and model.
Step 3 β Configure the Workspace
OpenClaw automatically creates ~/.openclaw/workspace/ and loads several Markdown files into every agent's context at session start. These are the primary way to configure agent behavior:
| File | Purpose |
|---|---|
| MEMORY.md | Loaded at session start β current projects, active tasks, key facts. Keep under ~200 lines. |
| AGENTS.md | Behavioral rules: how to route tasks, subagent patterns, session hygiene, media rules. |
| TOOLS.md | Tool usage guide: server SSH details, CLI syntax for MCP tools, search priority, etc. |
| SOUL.md | Agent personality and communication style. |
| IDENTITY.md | Agent identity: name, role, capabilities summary. |
| memory/ | Per-topic Markdown files loaded on demand: entities/, patterns/, lessons.md, etc. |
Each file is plain Markdown β edit them directly. Per-agent workspaces (e.g. ~/.openclaw/workspace-coding/) follow the same structure and override the defaults for that agent.
Step 4 β Connect Channels
OpenClaw connects to messaging platforms so you can talk to your agents from anywhere. Channels are configured in ~/.openclaw/openclaw.json. Supported channels include iMessage, Telegram, Discord, WhatsApp, and others.
Use openclaw onboard for the guided pairing flow, or consult the OpenClaw channel docs for manual configuration. Once a channel is paired, messages from your configured accounts will route to the gateway.
Step 5 β Add Claw-Stack Modules
Claw-Stack modules are separate projects you clone and configure alongside your running OpenClaw gateway. Each module integrates through one or more of OpenClaw's native extension points:
MCP Servers
Modules like agent-time-awareness and chrome-devtools-mcp expose MCP tools that the agent calls via mcporter. Start the MCP server, then configure it in TOOLS.md so the agent knows how to use it.
OpenClaw Skills
Skills are Markdown files in ~/.openclaw/workspace/skills/ that agents read before performing specialized tasks (image processing, document generation, etc.).
Cron Tasks
The OpenClaw cron system schedules recurring agent work β daily memory sync, backups, heartbeats. Modules like openclaw-backup add their backup script to the cron schedule.
OpenClaw Plugins
Modules like openclaw-security install as OpenClaw plugins that hook into before/after tool call events via the plugin system.
Each module's README documents its specific setup steps. Start with the ones most relevant to your use case:
| Module | Doc page |
|---|---|
| Multi-Agent Consensus (agent-meeting) | Modules β Multi-Agent Consensus Protocol |
| Smart Scheduler (agent-time-awareness) | Modules β Smart Scheduler & Deadline Watch |
| Web Automation (chrome-devtools-mcp) | Modules β Web Automation Operator |
| Intelligence Feed (info-pipeline) | Modules β Live Intelligence Feed |
| Backup (openclaw-backup) | Modules β Encrypted State Archive |
| Security (openclaw-security) | Modules β Governance & Security |
| Voice Interface (voice-call) | Modules β Executive Voice Interface |
Frequently Asked Questions
Do I need OpenClaw to use Claw-Stack?
Yes. Every Claw-Stack module is designed to work alongside a running OpenClaw gateway. The modules integrate through OpenClaw's extension points β MCP servers, skills, plugins, cron tasks, and workspace configuration files. None of them operate as a standalone system.
Where does OpenClaw store everything?
Everything lives under ~/.openclaw/: the gateway config (openclaw.json), the main workspace (workspace/), per-agent workspaces (workspace-coding/, etc.), agent-specific state (agents/), logs, media, cron schedules, and credentials.
Is there a web dashboard?
OpenClaw includes a built-in control UI that you can open from the gateway. It shows chat, session status, and configuration. This is separate from any Claw-Stack modules β it is part of OpenClaw itself. Check the OpenClaw docs for how to access it.
Authors: Qiushi Wu & Orange π