OpenClaw

Most AI coding tools assume you’re at a terminal, copying prompts and pasting responses. The agent exists in a browser tab, disconnected from everything else you use. You’re the integration layer.

OpenClaw takes a different approach: the agent lives in your infrastructure. It connects to your messaging apps, calendar, email, filesystem, and whatever else you wire up. Instead of context-switching to AI, the AI has context about you.

The core idea

OpenClaw is an open-source runtime that gives AI agents persistent access to your tools. It runs locally or on a VPS you control. Your data stays yours.

# Install
npm install -g openclaw

# Setup (interactive)
openclaw onboard

# Start the daemon
openclaw gateway start

Once running, you can reach your agent via Telegram, Discord, Signal, Slack, CLI, or web dashboard. One agent, multiple surfaces.

The philosophy: AI should adapt to your workflow, not the other way around. Persistent context beats fresh starts.

How it works

OpenClaw connects three things:

┌─────────────────────────────────────────────────────┐
│ Channels (how you talk to it)                       │
│ Telegram, Discord, Signal, Slack, CLI, Web         │
└─────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────┐
│ Gateway (the runtime)                               │
│ Routes messages, manages sessions, runs tools       │
└─────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────┐
│ Tools & Skills                                      │
│ Shell, browser, files, MCP servers, custom skills   │
└─────────────────────────────────────────────────────┘

The agent maintains context through workspace files:

AGENTS.md — How the agent should behave
SOUL.md — Personality and tone
USER.md — Context about who it’s helping
MEMORY.md — Long-term curated memory
memory/YYYY-MM-DD.md — Daily session notes

These files are injected into every conversation. The agent builds understanding over time, like a colleague who’s been paying attention.

Key capabilities

Proactive behavior

Unlike chat-only tools, OpenClaw can reach out to you:

# HEARTBEAT.md - checked every 30 minutes
- Check inbox for urgent emails
- Review calendar for upcoming meetings
- Monitor GitHub notifications

The agent polls on a schedule. When something needs attention, it messages you. No need to remember to ask.

Background tasks

Spawn sub-agents for parallel work:

You: "Research competitors while I focus on the pitch deck"
Agent: [spawns research sub-agent, continues helping with deck]
Sub-agent: [works independently, reports back when done]

Long-running tasks don’t block the conversation.

Tool integration

Built-in tools cover common needs:

Shell execution — Run commands, scripts, builds
Browser automation — Navigate, scrape, interact with web apps
File operations — Read, write, edit, organize
Web search — Research without leaving the conversation

Extend with skills for specific workflows:

# Install a skill
clawhub install trello

# Now the agent can manage your boards
"Move the 'API redesign' card to Done"

Multi-channel presence

Same agent, multiple ways to reach it:

Quick question via Telegram while mobile
Deep work session via CLI at your desk
Team collaboration via Discord or Slack
Dashboard for reviewing history

Context carries across channels. The agent knows what you discussed on Telegram when you continue on CLI.

When to use it

OpenClaw works well for:

Personal AI assistant — Inbox triage, calendar management, task tracking
Development workflows — Code review, PR monitoring, CI/CD notifications
Research and writing — Web research, drafting, editing with persistent context
Automation — Scheduled tasks, monitoring, proactive alerts

It’s less suited for:

One-off coding questions (just use ChatGPT)
Team-wide deployment (designed for personal/small team use)
Environments where local installation isn’t possible

Getting started

# Install globally
npm install -g openclaw

# Interactive setup - configures model, channels, workspace
openclaw onboard

# Start the gateway daemon
openclaw gateway start

# Check status
openclaw status

The onboarding wizard walks through:

Choosing an AI provider (OpenRouter, Anthropic, OpenAI, etc.)
Setting up channels (Telegram bot, Discord, etc.)
Configuring your workspace

Resources

GitHub Repository — Source code
Documentation — Full docs
ClawHub — Find and share skills
Discord Community — Support and discussion

OpenClaw is open source and actively developed. The project welcomes contributions.

Note: OpenClaw was originally called “ClawdBot”, then “MoltBot”, before landing on “OpenClaw”.