AI agents are everywhere. Managing them is still a mess.

You've got an API key for Claude. Maybe one for GPT-4o too. You've seen demos of AI bots that respond on Telegram, Discord, WhatsApp - and you want to build something like that. But every tutorial sends you down a different rabbit hole: custom servers, bespoke webhook handlers, brittle Python scripts that break whenever you add a second agent.

OpenClaw was built to solve exactly this. It's the open-source engine that runs your AI agents - handling the plumbing so you can focus on what the agents actually do.

What is OpenClaw?

OpenClaw is an open-source AI agent gateway. Think of it as a runtime - a persistent process that sits on your machine (or server) and orchestrates one or more AI agents, each connected to messaging platforms, each with their own model, memory, and behaviour.

A single OpenClaw gateway can run multiple agents simultaneously. Each agent is independently configured: it can use a different AI model, connect to different channels, have different skills and tools, and respond to different triggers. They can also talk to each other - spawning sub-agents, delegating tasks, and collaborating on complex workflows.

Under the hood, OpenClaw handles:

Channel routing - incoming messages from Telegram, Discord, WhatsApp, Slack, Signal, iMessage, and 8 more platforms are normalised and delivered to the right agent
Session management - conversation history, context windows, and compaction are handled automatically
Model abstraction - swap between Claude, GPT-4o, Gemini, Mistral, Ollama, and others without changing your agent's logic
Tool execution - agents can run shell commands, read and write files, browse the web, call APIs, and more
Automation - heartbeat check-ins and cron jobs keep agents working even when no one is talking to them

Key concepts

Agents

An agent is the core unit in OpenClaw. Each agent has an identity (name, persona, instructions), a configured AI model, a set of tools it's allowed to use, and one or more channels it listens on. Agents are defined by a set of plain text files - SOUL.md, IDENTITY.md, AGENTS.md, USER.md - that shape how they think and respond. This makes them easy to version, back up, and share.

Channels

Channels are the messaging platforms your agents connect to. OpenClaw currently supports Telegram, Discord, WhatsApp, Slack, Signal, iMessage, Google Chat, IRC, and more. Each agent can connect to multiple channels at once. You can also set wakewords so an agent only responds when specifically addressed in a group chat.

Sessions

Every conversation is a session. OpenClaw tracks session history, compacts context automatically when it gets long, and lets you inspect, search, and replay past sessions. Sessions are local and private by default - nothing is sent to a third party beyond the model API call itself.

Skills

Skills are reusable instruction sets that extend what an agent can do. A skill might teach an agent how to use a specific tool, follow a specific workflow, or behave differently in certain contexts. Skills live in plain text files and are loaded into the agent's context at runtime.

Heartbeat & Cron

Agents don't have to be reactive. With heartbeat, you can schedule periodic check-ins - the agent wakes up on a timer, reviews its task list or monitors a feed, and acts if needed. Cron jobs work similarly, triggering specific prompts on a schedule. This is how you build agents that work autonomously in the background.

Getting Started with OpenClaw & AI Agents

Select which method you prefer for setup, visual or terminal:

Getting started (ClawManager + OpenClaw)

If you prefer a visual setup over the terminal, ClawManager is the fastest way to get OpenClaw running. A setup wizard walks you through everything on first launch.

1. Download ClawManager

Head to clawmanager.ai and download the app for macOS, Linux, or Windows. Install it and open it. The setup wizard starts automatically on first launch.

2. Install OpenClaw

If OpenClaw is not on your machine yet, the wizard will offer to install it for you. Click Install OpenClaw and ClawManager handles it with no terminal and no npm required. If OpenClaw is already installed it detects it automatically and moves on.

3. Add your AI provider key

The wizard prompts you to paste an API key. Anthropic, OpenAI, Google, Mistral, and Ollama are all supported. ClawManager writes the key to your OpenClaw config so your agent can start making model calls immediately. You only need one key to begin.

4. Start your gateway

Click Start Gateway. ClawManager launches your OpenClaw gateway in the background and keeps it running. The dashboard shows your agent status, active sessions, and live output in real time without needing a terminal window open.

5. Connect a channel

Open the Connections panel and pick a platform. Telegram is the easiest starting point: create a bot via @BotFather, paste the token into ClawManager, and hit save. The gateway restarts with the channel active and your agent is now live on Telegram.

6. Customise your agent

The Agent tab in ClawManager lets you edit your agent SOUL.md, USER.md, and other workspace files directly in the app. No file manager or text editor needed. Changes are picked up on the next message.

Getting started (Terminal) - OpenClaw

OpenClaw runs on macOS, Linux, and Windows. The quickest way to get going if you are comfortable with a Terminal:

1. Install OpenClaw

Download and run the installer from openclaw.ai, or install via the command line:

npm install -g openclaw

Once installed, initialise your first workspace:

openclaw init

This creates a ~/.openclaw directory with the default config and a starter agent.

2. Add your AI provider keys

OpenClaw supports every major provider. Add your keys through the config or by editing ~/.openclaw/config.json directly. You only need the key for the model you want to use - you can start with a free-tier Anthropic or Google key.

3. Start the gateway

openclaw gateway start

Your gateway is now running. You can talk to your first agent immediately through the built-in CLI chat:

openclaw chat

4. Connect a channel

To connect Telegram, create a bot via @BotFather, copy the token, and add it to your agent's channel config. OpenClaw handles the webhook setup automatically. The same process applies to Discord (bot token), Slack (app credentials), and every other supported platform.

5. Customise your agent

Edit your agent's SOUL.md to define its personality and working style. Edit USER.md to tell it about you. Edit AGENTS.md to define how it should behave with other agents. These are just Markdown files - edit them in any text editor and the agent picks up changes on the next message.

Where ClawManager fits in

Once you have more than one agent - or once you want to manage heartbeats, inspect sessions, configure models, or back up your setup - the command line starts to feel like a lot. ClawManager is a free desktop app that puts a UI on top of your OpenClaw gateway. It doesn't change how OpenClaw works; it just makes it easier to see what's happening, configure things visually, and manage multiple agents side by side.

If you're comfortable in the terminal, you don't need it. But most people find it saves a lot of time once their setup grows past one or two agents.

What can you do with your Agent(s) ?

The range of what you can build with OpenClaw is genuinely wide. Some examples from the community:

A personal assistant that lives in your Telegram, reads your emails, manages your calendar, and briefs you every morning
A team of specialised agents - a researcher, a writer, a coder - that collaborate on long-running projects
A customer support bot for a Discord server that answers questions, escalates edge cases, and logs conversations
A monitoring agent that watches a feed, a GitHub repo, or a website and alerts you when something changes
A coding assistant that runs directly on your machine with full file access, no subscription required

The common thread: agents that actually do things, connected to the platforms you already use, running on infrastructure you control.

Next steps

The best way to understand OpenClaw is to run it. Install it, get your first agent talking on Telegram, and go from there. The documentation covers every channel, every tool, and every config option in detail.

If you want a visual starting point, download ClawManager - the setup wizard will walk you through installing OpenClaw, configuring your first agent, and connecting your first channel without touching a config file.