Languages: English | 简体中文 | हिन्दी | Español | العربية | Français | বাংলা | Português | Русский | Bahasa Indonesia
Developer-first curated guide to OpenClaw. This README prioritizes official docs, practical setup order, security baselines, and durable ecosystem links over hype, pricing tables, and fast-staling marketing claims.
- Product Overview
- Quick Start & Setup Methods
- Install Paths
- Core Concepts & Commands
- Channels: What to Connect First
- Deployment Choices
- Security Baseline
- Official Docs by Goal
- Curated Ecosystem Links
- Common Mistakes
- Contributing
OpenClaw is a personal AI assistant you run on your own devices. At the center is the Gateway, which acts as the control plane for agents, channels, sessions, tools, dashboard/web UI, and remote access. A strong way to think about it is: OpenClaw is not "just a bot" - it is a local-first assistant runtime that can live inside the chat surfaces you already use.
- A local Gateway with CLI, dashboard, WebChat, and remote access options
- Multi-channel messaging across surfaces like WhatsApp, Telegram, Slack, Discord, WebChat, and more
- Multiple agents with separate workspaces, sessions, bindings, and model policies
- Flexible model setup across hosted providers, OAuth-based auth, API keys, and local-model paths
- Skills, browser automation, nodes, cron jobs, and other tool-driven workflows
- A personal-assistant trust model, not a hostile multi-tenant SaaS security model
| Layer | Purpose |
|---|---|
| Gateway | Central control plane - message routing, sessions, plugins, tool execution policy |
| Channels | Adapters normalizing Telegram/WhatsApp/Discord/iMessage into standard message shapes |
| Routing + Sessions | Determines which agent handles specific conversations |
| Agent Runtime | Processes context, calls model providers, streams responses, requests tools |
| Tools | Capabilities - web fetch, browser control, command execution, device pairing |
| Surfaces | Interaction points - chat apps, web dashboard, macOS menu bar, Live Canvas |
| Date | Name | Reason |
|---|---|---|
| Nov 2025 | Clawdbot | Original name at launch |
| Jan 27, 2026 | Moltbot | Anthropic trademark complaint (similarity to "Claude") |
| Jan 30, 2026 | OpenClaw | Final rebrand, community vote |
Good fit if you want to:
- run a personal assistant on your own laptop, desktop, home server, or VPS
- keep long-lived chat history, files, and automation inside one workspace
- connect the same assistant to multiple messaging channels you already use
- experiment with tools, skills, browser automation, or multi-agent routing
Not the best fit if you need:
- hostile multi-tenant isolation on one shared gateway
- a zero-maintenance consumer app with no setup or operational ownership
- enterprise compliance, SLAs, and vendor support out of the box
If you only do one thing: get the local dashboard working before you connect any external channel.
| Item | Recommendation |
|---|---|
| Runtime | Node 24 recommended; Node 22.16+ minimum |
| OS | macOS or Linux; Windows via WSL2 is the recommended Windows path |
| First interface | Dashboard / WebChat |
| First setup flow | openclaw onboard |
npm install -g openclaw@latest
openclaw onboard --install-daemon
openclaw dashboardcurl -fsSL https://openclaw.ai/install.sh | bash
openclaw onboard --install-daemonnpm install -g openclaw@latest
# or
pnpm add -g openclaw@latestgit clone https://github.com/openclaw/openclaw.git
cd openclaw
./docker-setup.sh| Platform | Link | Notes |
|---|---|---|
| OpenClaw Launch | 1-Click Deploy | Official, fastest to get started |
| Railway | Template | Web-based setup wizard |
| DigitalOcean | 1-Click Deploy | Security-hardened, pre-configured |
| Render | render.yaml | Infrastructure as Code |
| SimpleClaw | simpleclaw.com | Deploy in under 1 minute |
| Zeabur | Template | One-click Docker deploy |
| Northflank | Stack | No server-side terminal needed |
| Lightning.AI | Environment | Browser-based, no local hardware |
| Coolify | Template | Self-hosted PaaS template |
| Elestio | Open Source | Fully managed in under 3 minutes |
| # | Product | Website | Type | Self-Host | Messaging Platforms |
|---|---|---|---|---|---|
| 1 | Hermes Agent | hermesagent.studio | Autonomous agent + messaging hub | Yes | WhatsApp, Telegram, Slack, Discord, and more |
| 2 | Multica | multica.uk | Multi-channel AI automation | Yes | Multi-platform |
| 3 | GenericAgent | genericagent.org | Agent workspace with browser, terminal, filesystem, and memory control | Not specified | Web workspace |
| 4 | Ruflo AI | ruflo.online | Hosted Ruflo-style multi-agent workspace launch layer | No | Web workspace |
| 5 | AutoGPT | agpt.co | Autonomous task agent | Yes | API / web UI |
| 6 | LangChain | langchain.com | LLM orchestration framework | Yes | Any (via custom integrations) |
| 7 | n8n | n8n.io | Workflow automation + AI nodes | Yes | Slack, Telegram, Discord, and 400+ apps |
| 8 | AgentGPT | agentgpt.reworkd.ai | Browser-based autonomous agent | Yes | Web UI |
| 9 | CrewAI | crewai.com | Multi-agent role collaboration | Yes | API / custom integrations |
| 10 | SuperAGI | superagi.com | Autonomous agent infrastructure | Yes | Slack, Email, API |
- Run
openclaw onboardand pick QuickStart unless you already know the exact config you want. - Choose one model provider and one default model.
- Keep the gateway local-only on the first run.
- Open the dashboard and verify you can chat without any external channel.
- Add exactly one channel after that.
- Run
openclaw doctorandopenclaw security auditbefore exposing anything remotely.
- Dashboard or WebChat first
- One model provider
- One channel
- Security audit
- Skills and browser automation
- Remote access, VPS, or always-on deployment
| Path | Best For | Why It Helps | Link |
|---|---|---|---|
| CLI onboarding | Most developers | Fastest official path; configures gateway, workspace, channels, daemon, and skills | Getting Started / Onboarding |
| Docker | VPS, cleaner runtime isolation | Easier to contain the runtime and redeploy cleanly | Docker |
| From source | Contributors and debuggers | Full pnpm build and watch workflow from the main repo |
openclaw/openclaw |
| Nix | Reproducible environments | Declarative setup and updates | nix-openclaw |
| macOS app path | Apple-device-heavy setups | Most useful if you specifically care about voice, canvas, and native platform features | macOS Platform Docs |
For a normal developer, the default answer is still: install from npm, run onboarding, and validate everything in the dashboard before doing anything more ambitious.
| Concept | What It Means | Why It Matters |
|---|---|---|
| Gateway | The local control plane | Owns config, sessions, channels, tools, dashboard, and auth |
| Agent | A logical assistant identity | Lets you separate use cases, workspaces, and model policies |
| Workspace | Files and prompts for an agent | Where skills, notes, and task artifacts live |
| Channel | A messaging integration | Determines how users reach the assistant |
| Session | A conversation context | Controls memory continuity and routing behavior |
| Skill | A packaged capability | Reusable prompt + files + scripts for common tasks |
| Tool profile | Allowed capability set | One of the biggest security and behavior levers |
| Model policy | Primary model plus fallbacks | Controls quality, cost, and tool-safety posture |
| Task | Command |
|---|---|
| Run guided setup | openclaw onboard |
| Reconfigure later | openclaw configure |
| Open browser dashboard | openclaw dashboard |
| Inspect health and migrations | openclaw doctor |
| Audit security posture | openclaw security audit |
| Deep security audit | openclaw security audit --deep |
| Add another agent | openclaw agents add <name> |
| Inspect config | openclaw config get <path> |
| Update config | openclaw config set <path> <value> |
| Check model auth or status | openclaw models status |
| Quick local prompt | openclaw agent --message "Ship checklist" |
- Use onboarding and
openclaw configurebefore hand-editing a lot of JSON. - Create a second agent early if you want to separate personal use from work automation.
- Check
openclaw models statuswhenever a provider feels "configured but not working."
Do not connect five channels on day one. Pick the surface that matches your real goal and setup tolerance.
| Channel | Best First Use | Notes | Docs |
|---|---|---|---|
| WebChat | Fastest smoke test | No third-party bot setup; ideal for first validation | WebChat |
| Telegram | Easiest external bot path | Good first real channel for developers | Telegram |
| Daily personal assistant | Great for real-world usage, but keep DM policies tight | ||
| Slack | Team or internal workflow | Learn trust boundaries before broad workspace access | Slack |
| Discord | Private server or community experiments | Tighten DM and group rules before opening it up | Discord |
| Signal | Privacy-oriented personal usage | Extra dependency, worth it if Signal is where you already live | Signal |
- Want the fastest success path: start with WebChat or Telegram.
- Want a real daily-driver: move to WhatsApp after the dashboard works.
- Want team usage: learn security and agent separation first, then use Slack or Discord.
| Scenario | Recommended Path | Why |
|---|---|---|
| First evaluation | Local CLI onboarding + dashboard | Lowest complexity and easiest debugging |
| Personal always-on assistant | Home server / Mac mini / Linux box + daemon | Stable daily-driver setup with local ownership |
| Cheap VPS | Docker | Cleaner host boundary and simpler redeploy story |
| Reproducible personal infra | Nix | Declarative config and repeatable installs |
| Remote access | Tailscale or SSH tunnel | Safer than exposing the gateway blindly to the internet |
| Multiple trust boundaries | Separate gateways, and ideally separate OS users or hosts | OpenClaw is not designed as hostile multi-tenant isolation on one shared gateway |
Remote access is something to add after local setup, not before. The common failure mode is exposing a powerful assistant before pairing rules, allowlists, tool profiles, and agent boundaries are thought through.
This is the section many "awesome" lists skip and developers actually need.
- Treat inbound DMs, web content, shared chats, and uploaded files as untrusted input.
- Keep DM access in pairing or strict allowlist mode while you are still learning the system.
- If more than one person can DM the bot, use
session.dmScope: "per-channel-peer"or a stricter equivalent. - Keep group channels mention-gated unless you have a specific reason not to.
- Start with a restrictive tool profile and only widen access when you know exactly why.
- For shared or non-main sessions, prefer sandboxing rather than trusting prompt discipline alone.
- Keep the gateway local-only until you intentionally configure remote access and auth.
- Run
openclaw security auditregularly, and use--deepwhen you are about to widen exposure. - Separate personal and company trust boundaries into different agents, and preferably different gateways or hosts.
| Goal | Best Link |
|---|---|
| First install | Getting Started |
| Guided setup | Onboarding (CLI) |
| Browser-first testing | Dashboard |
| Configuration | Configuration |
| Model setup | Models CLI |
| Channels overview | Channels |
| Skills | Skills |
| Browser automation | Browser Tool |
| Docker install | Docker |
| Security hardening | Security |
| Remote exposure | Remote Access |
| Remote access via Tailscale | Tailscale |
| Troubleshooting | FAQ |
- Getting Started
- Onboarding
- Dashboard
- Configuration
- One channel doc
- Security
- Skills, browser automation, and remote access
Browse these after you have one working install. They are much more useful once the official onboarding path already makes sense to you.
- OpenHuman Online - inspectable AI memory tree planning for teams running assistants with human-readable source context.
| Resource | Why It Matters |
|---|---|
| openclaw/openclaw | Main codebase and the authoritative README |
| docs.openclaw.ai | Official documentation hub |
| openclaw/clawhub | Official skill directory and package catalog |
| openclaw/nix-openclaw | Nix packaging for reproducible installs |
| Resource | Why It Matters |
|---|---|
| essamamdani/openclaw-coolify | Coolify template for self-hosted PaaS users |
| serhanekicii/openclaw-helm | Helm chart if your deployment world is Kubernetes |
| lunarpulse/openclaw-mcp-plugin | MCP-oriented integration starting point |
| centminmod/explain-openclaw | Deep technical documentation and deployment or security notes |
- Reading massive hosting tables before you have a local chat working
- Connecting too many channels at once, then debugging setup noise instead of one clear path
- Treating a shared tool-enabled agent as if it were a safe multi-tenant app
- Exposing the gateway remotely before pairing, allowlists, and auth are tested locally
- Mixing personal and company trust boundaries inside one powerful assistant runtime
- Expecting strong local-model performance on underpowered hardware without planning for it
- Hand-editing lots of config before onboarding has created a sane baseline
See CONTRIBUTING.md.
When adding links, prefer resources that help a normal developer do one of these jobs well:
- install OpenClaw
- configure it correctly
- secure it sensibly
- debug it faster
- extend it with a real development payoff
This list is released under CC0 1.0 Universal (Public Domain).