Open Source · Local-First · Model-Agnostic

Your AI Agent.
Running on Your Terms.

OpenClaw is an autonomous AI agent platform that runs on your hardware, connects to your messaging apps, and executes real tasks — without sending your data anywhere.

Terminal
$ openclaw onboard --install-daemon
✓ Node.js v24.0.0 detected
✓ Gateway started on port 18789
? Select AI provider: Anthropic Claude
✓ API key configured
? Connect a channel: Telegram
✓ Channel paired successfully
🚀 OpenClaw is running. Message your agent!
1 command to install
5+ messaging apps
100% runs on your hardware
extensible skills

Works with your stack

🤖 Claude
🧠 GPT-4o
🔵 DeepSeek
🦙 Ollama (local)
✈️ Telegram
💬 WhatsApp
🎮 Discord
💼 Slack

Founders are stuck in a broken loop

Cloud AI tools are fragmented. Context doesn't persist. Your data leaves your machine. Nothing actually does the work — it just suggests it.

❌ Without OpenClaw
  • 5 tabs, 3 subscriptions, zero continuity
  • Copy-paste between tools all day
  • Data uploaded to servers you don't control
  • Tasks start but never finish autonomously
  • Setup takes days. Value takes weeks.
  • Context is lost after every conversation
VS
✅ With OpenClaw
  • One agent. One message. Work gets done.
  • Runs on your hardware, always on
  • Your data never leaves your machine
  • Autonomous task execution via Skills
  • Setup in 5 minutes. Value in minutes.
  • Persistent memory and context across sessions
🔒

Private by design

Runs entirely on your infrastructure. API keys and files never leave your machine. No cloud dependency, no data risk.

Actually executes

Not just a chatbot. OpenClaw reads files, runs shell commands, browses the web, calls APIs. It does things.

💬

Works in your inbox

Chat with it via Telegram, WhatsApp, or Discord. No new UI to learn. Your existing apps, now powered by AI.

🧩

Extensible Skills

Define custom skills as Markdown files. The agent learns new capabilities without code changes.

🔄

Always-on automation

Heartbeat scheduler lets your agent act proactively — daily reports, monitoring, scheduled outreach — without you asking.

🎛️

Model-agnostic

Claude, GPT-4o, DeepSeek, or local models via Ollama. You choose the model per task or per agent.

How OpenClaw works

From install to first task in minutes. Here's exactly what happens under the hood.

01

Install the Gateway

One command downloads OpenClaw and starts the local Gateway — your agent's control plane that runs as a background daemon.

openclaw onboard --install-daemon
02

Connect an AI model

Pick your preferred LLM provider — Anthropic, OpenAI, DeepSeek, or run a local model with Ollama. Your keys stay on-device.

Claude GPT-4o DeepSeek + Custom
03

Pair a messaging channel

Connect Telegram (BotFather token), WhatsApp (QR code), Discord, or Slack. Instantly talk to your agent from any device.

✈️ Telegram 💬 WhatsApp 🎮 Discord
04

Define skills & workspace

Create a SOUL.md to set the agent's persona and add Skills markdown files to teach it custom capabilities for your workflow.

workspace/
SOUL.md — persona & instructions
AGENTS.md — agent config
skills/
research.md
outreach.md
05

Message. Delegate. Ship.

Just send a message. OpenClaw picks the right skill, executes the task, and sends back the result. It works while you sleep.

Research competitors for our CRM pitch deck
OpenClaw On it. Scanning 12 sources across Product Hunt, G2, and Crunchbase…
✓ Done in 4 minutes. Summary attached.

Skills make OpenClaw unstoppable

Skills are lightweight Markdown files that define new capabilities. No code required. Drop a file in your workspace and your agent instantly knows what to do.

🔍

Research

Deep-dive competitor analysis, market sizing, technical documentation scanning. Synthesizes sources automatically.

Built-in
📁

File Operations

Read, write, organize, and search across local and cloud files. Process CSVs, PDFs, and project folders.

Built-in
⚙️

Shell Execution

Run any CLI command on your machine. Build, test, deploy, or monitor servers directly from your chat.

Built-in
🌐

Web Browsing

Browse, scrape, and extract information from any webpage. No manual copy-paste. Just ask and get results.

Built-in

Heartbeat Scheduler

Define recurring tasks. Daily standup summaries, weekly reports, system health checks — all autonomous.

Built-in
🔗

API Integration

Call any REST API. Connect CRMs, analytics tools, databases, or notifications services via custom skills.

Custom
📊

CRM & Outreach

Automatically update records, draft personalized emails, and manage lead pipelines without lifting a finger.

Custom
💻

Code Review

Review pull requests, scan logs, run tests, and create GitHub issues. Built for technical operators.

Custom

Create your own skill in seconds

Skills are just Markdown files with structured frontmatter. No SDK. No boilerplate.

View Skill Docs →

What founders actually use it for

Not theoretical demos. These are real workflows from founders running OpenClaw in production today.

Your personal founder AI, always on

Run OpenClaw on a Mac Mini or VPS. Connect Telegram. Now you have a 24/7 executive assistant that never sleeps, never bills by the hour, and works exactly how you train it to.

  • Daily standup summaries from Slack and email
  • Competitor monitoring and weekly briefings
  • Burn rate and runway calculations from your spreadsheets
  • Meeting prep from calendar invites
  • Draft investor updates from your notes
Explore founder workflows →
What's our burn rate this week?
Reading expenses from Notion... Calculated. Burn: $18,400/wk. Runway: 14.2 months at current rate. 📊
Draft an update email for investors
Draft ready. Sent to your drafts folder. Want me to also schedule for Tuesday morning?

Automate your dev operations

Connect OpenClaw to your GitHub, deployment scripts, and monitoring stack. Let it watch your CI/CD pipeline and tell you when things break — or fix them automatically.

  • Review PRs and catch common issues
  • Monitor Sentry and auto-create tickets
  • Run test suites via shell commands
  • Deploy to staging on command
  • Daily log summaries for on-call handoffs
Explore dev workflows →
🚨 Sentry alert: 142 errors from payment module in last 10 min
Investigate and create a GitHub issue
Root cause: null pointer on expired session token. Issue #847 created. PR branch opened. Fix ready for review.

Deep research, synthesized fast

Stop spending hours on competitive research. OpenClaw can scan dozens of sources, pull structured data, and deliver a clean briefing — while you focus on building.

  • Competitor landscape analysis
  • Market sizing from public data
  • Technical documentation synthesis
  • Industry news monitoring and digest
  • Investor research before pitches
Explore research workflows →
Give me a competitor analysis for our product hunt launch next week
Scanning Product Hunt, G2, Capterra, 8 subreddits, and 3 industry newsletters... Done. 14 competitors mapped. Top 3 with feature gaps highlighted. PDF ready.

Autonomous sales operations

Build an SDR agent that finds leads, drafts personalized outreach, follows up on schedule, and logs everything to your CRM — without hiring anyone.

  • Lead identification from target criteria
  • Personalized cold email drafting
  • Follow-up sequence management
  • CRM updates and pipeline maintenance
  • Reply sentiment analysis and prioritization
Explore sales workflows →
Find 20 fintech startups raising Series A and draft outreach
Found 23 matching companies on Crunchbase. 20 outreach emails drafted with personalized hooks. CRM updated. Want to review before sending?

What trips people up during setup

Most setup failures happen in the first 30 minutes. Here's what to watch for.

⚠️

Wrong Node.js version

OpenClaw requires Node.js v22.14+ (v24 recommended). Older versions cause silent failures during gateway startup.

Fix: node --version → upgrade if below v22.14
⚠️

Running on your main account

Never connect OpenClaw to your primary personal Telegram, WhatsApp, or GitHub. Use dedicated bot accounts to isolate access.

Fix: Create isolated bot/service accounts before onboarding
⚠️

Gateway not running

If your agent stops responding, the Gateway daemon may have crashed. Always verify it's running before debugging further.

Fix: openclaw gateway status → restart if needed
⚠️

Running on a primary laptop

OpenClaw is always-on. Running it on your daily laptop means it stops when you close the lid. Use a Mac Mini, VPS, or home server.

Fix: Deploy on a dedicated always-on machine for 24/7 operation
⚠️

Discord Intent not enabled

Discord requires "Message Content Intent" enabled in the Developer Portal. Missing this causes bot silence — no errors thrown.

Fix: Enable Message Content Intent in Discord Developer Portal
⚠️

Skills not discovered

Skills must live in your workspace skills/ folder and have correct frontmatter. Misconfigured files are silently ignored.

Fix: Run openclaw doctor to validate your workspace

From zero to running in 5 minutes

We've documented every step. No fluff, no gaps. Just a clear path to your first working agent.

1
Install Node.js v24

nodejs.org or nvm install 24

2
Run the installer

openclaw onboard --install-daemon

3
Connect a channel

Follow prompts to pair Telegram or WhatsApp

4
Send your first message

Your agent is live. Start delegating.

Full Setup Guide →