One Agent. Twelve Messaging Apps. Zero Dashboards.
You already have too many dashboards.
Your project manager has one. Your CRM has one. Your hosting provider has one. Your analytics tool has one. You probably have 15+ browser tabs open right now, each one a different dashboard demanding your attention.
Now imagine adding another one just to talk to your AI agent.
That's what most agent frameworks expect. A web UI. A CLI. Another browser tab.
OpenClaw takes the opposite approach.
Your agent lives in the messaging apps you already have open. WhatsApp. Telegram. Discord. Slack. The apps on your phone, on your desktop, on your watch.
No new dashboard. No new tab. No new app to learn.
What "Multi-Channel" Actually Means
OpenClaw runs a single Gateway process — one server on port 18789. That gateway connects to multiple messaging platforms simultaneously.
Here's what's supported natively (built into the core):
- WhatsApp — via the Baileys library (Web protocol). QR pairing. Per-contact allowlists. Group chat support.
- Telegram — via the grammY library (Bot API). Fastest setup. Create a bot with @BotFather, paste the token, done.
- Discord — Bot API + Gateway. Server channels and DMs. Slash command support.
- Slack — via Bolt SDK. Workspace app integration.
- Signal — via signal-cli. Privacy-focused messaging.
- iMessage — via BlueBubbles (macOS). Send and receive from your Apple ID.
- Google Chat — HTTP webhook API.
- IRC — Classic channel and DM support.
- WebChat — Built-in browser interface over WebSocket.
And through the plugin system (installable separately):
- Microsoft Teams (Bot Framework)
- LINE (Messaging API)
- Matrix (Matrix protocol)
- Mattermost (Bot API + WebSocket)
- Feishu/Lark (WebSocket)
- Nextcloud Talk
- Synology Chat
- Nostr (NIP-04 DMs)
- Twitch (IRC)
- Zalo
One agent. All of these. At the same time.
Why This Architecture Matters
Most AI agent tools force you into their interface. You open their app, type your request, wait for a response.
OpenClaw flips this. The agent meets you where you are.
Morning commute? Text your agent on WhatsApp. Ask for your morning brief. It arrives as a message you can read on your phone.
At your desk? Message via Slack. Your team sees the exchange. No context lost.
Late at night? Telegram DM. Quick question, quick answer.
Team standup? Your agent posts the daily summary in a Discord channel. Everyone sees it.
The practical benefit: you stop context-switching.
You don't open Dashboard A to check one thing, then open Dashboard B to check another. You stay in the app you're already using and let the agent bring the information to you.
How Channels Actually Work (The Real Config)
Channels are configured in openclaw.json. Each channel type has its own section with connection credentials and policy settings.
Here's what a real multi-channel config looks like:
{
channels: {
telegram: {
enabled: true,
botToken: "123456:ABC-your-bot-token",
dmPolicy: "pairing", // Unknown senders must pair first
groupPolicy: "allowlist" // Only approved groups
},
whatsapp: {
enabled: true,
dmPolicy: "pairing",
allowFrom: ["+15551234567"],
groupAllowFrom: ["+15551234567"]
},
discord: {
enabled: true,
botToken: "your-discord-bot-token"
},
slack: {
enabled: true
// Bolt SDK config
}
}
}
Each channel has independent security policies:
dmPolicy controls who can DM your agent:
pairing— Unknown senders receive a one-time pairing code. They must complete pairing before the agent responds. (Default)allowlist— Only pre-approved numbers/users can message.open— Anyone can message. (Not recommended for production.)disabled— DMs are off entirely.
groupPolicy controls group chat behavior:
- Agent responds to mentions in approved groups.
- Group chats get isolated sessions automatically — conversations don't bleed between channels.
The pairing flow is OpenClaw's answer to spam and unauthorized access. When someone unknown messages your agent, they get a code. You approve or reject the pairing. No approval, no access.
Session Isolation: Conversations Don't Bleed
Here's a detail that matters more than it sounds.
When you message your agent on Telegram about Project A, and then message on Discord about Project B, those are separate sessions.
OpenClaw routes sessions based on channel and sender. Direct chats share a "main" session per user. Group chats get isolated sessions automatically.
This means your team member's Discord conversation about the marketing plan doesn't interfere with your Telegram conversation about the deployment pipeline.
Queue modes control what happens when the agent is already working on a task:
- steer — Your new message gets injected into the current task. Useful for corrections: "Actually, make that a table instead of a list."
- followup — Your message queues until the current task finishes. Then it starts a new task.
- collect — Messages batch up. The agent processes them all at once in the next turn.
These aren't abstract features. They solve the real problem of asynchronous communication with an AI that can only do one thing at a time per session.
Multi-Agent Routing
One gateway can run multiple agents. Each agent has its own ID, workspace, model configuration, and personality.
Messages route to the right agent based on:
- Session filters — route by channel, sender, or topic
- Agent ID — explicitly address a specific agent
- Sender allowlists — certain users always route to a specific agent
Practical example: You have three agents on one gateway.
Agent 1 ("ops") monitors infrastructure alerts in a Discord channel. It uses Claude for reasoning about server issues.
Agent 2 ("research") handles research tasks via Telegram. It uses a local Ollama model for privacy.
Agent 3 ("support") responds to customer questions in Slack. It uses GPT-4 for consistent, friendly responses.
Same gateway. Different agents. Different models. Different channels. Each one specialized.
Clawctl's pricing scales with this: Starter supports 1 agent, Team supports 5, Business supports 25.
The Security Layer for Multi-Channel
When your agent is reachable from multiple messaging apps, security matters even more.
Every channel is a potential entry point. Every messaging app is a surface that someone could use to send your agent malicious instructions.
OpenClaw handles this with per-channel policies (pairing, allowlists) and per-channel session isolation.
For production deployments, Clawctl adds:
- Gateway authentication — 256-bit token. Every API call authenticated.
- Human-in-the-loop approvals — Risky actions from any channel require your explicit approval. The agent pauses, sends you a notification, waits for yes/no.
- Audit trail — Every message from every channel logged. Searchable by timestamp, channel, sender, action type.
- Kill switch — One click stops the agent across all channels. Not per-channel. All of them. Immediately.
- Encrypted secrets — API keys encrypted at rest. Doesn't matter which channel triggered the action — credentials are never exposed.
The channels are the interface. The security wraps around all of them equally.
How to Set Up Multi-Channel in Practice
Fastest channel to set up: Telegram. Create a bot with @BotFather on Telegram. Copy the token. Paste it in your config. Done in under 2 minutes.
Most useful for teams: Discord or Slack. Daily briefs, alerts, and team-visible conversations. 10 minutes to configure.
Most personal: WhatsApp. Text your agent like you'd text a colleague. QR pairing takes 5 minutes.
Start with one channel. Add more as you find uses for them.
If you're deploying to production, use Clawctl. $49/month. Deploys in 60 seconds. All channels secured with gateway auth, human approvals, and audit logging from day one.
Your agent doesn't need another dashboard.
It just needs to be where you already are.