Claude Code channels vs OpenClaw: a dev tool isn't a business agent

By Linas Valiukas · March 20, 2026

Anthropic shipped a feature called channels for Claude Code. It lets you pipe Telegram and Discord messages into a running terminal session. Claude reads the message, does something, and replies. Every AI newsletter will describe this as "Claude can be your Telegram bot now."

That framing skips a few things.

What channels actually are

A channel is an MCP server that pushes events into your Claude Code CLI session. Right now, that means Telegram and Discord. You create a bot on the platform side, install a plugin in Claude Code, pair your account, and start the session with a --channels flag.

When someone messages your bot, the message shows up in your terminal. Claude processes it and calls a reply tool. The response appears on Telegram or Discord. It works. It's tidy engineering.

But the key sentence in Anthropic's own documentation is this: "Events only arrive while the session is open."

Close the laptop, kill the bot

Claude Code is a CLI tool. It runs in a terminal. When that terminal closes, your "Telegram bot" stops responding. No session, no channel, no replies.

Anthropic's suggested workaround: "for an always-on setup you run Claude in a background process or persistent terminal." So to keep a messaging bot running 24/7, you need a server, a process manager, and the knowledge to keep both alive through reboots, crashes, and updates. You're self-hosting. The thing you were trying to avoid.

OpenClaw was built to run continuously. It's a server application. Conversation state persists across restarts. On TryOpenClaw.ai, uptime is our problem, not yours.

No memory, no schedule, no proactive behavior

Claude Code sessions are stateless by design. Each session starts fresh. Your bot doesn't remember yesterday's conversation. It doesn't know who messaged last week. There's no persistent user context, no long-term memory, no relationship building across interactions.

OpenClaw maintains full conversation history per user. Your agent knows who you are, what you asked last Tuesday, and what follow-up you're waiting on.

There's also no proactive behavior. Claude Code channels are reactive — a message comes in, a response goes out. That's it. No morning briefings sent at 7 AM. No "your competitor just changed their pricing" alerts. No follow-up messages after 48 hours of silence. No scheduled reports. An OpenClaw agent acts on its own. A Claude Code channel waits to be spoken to.

Two platforms, no WhatsApp

The research preview supports Telegram and Discord. Those are fine platforms. They're also not where most business conversations happen.

WhatsApp has 2+ billion users. It's the default messaging app for business in most of the world. Connecting an AI agent to WhatsApp is the most common request we see. Claude Code channels can't do it. Building your own WhatsApp channel plugin means dealing with the Meta Business API, webhook verification, phone number registration, and WhatsApp's messaging policies — all on your own.

OpenClaw supports WhatsApp, Telegram, Discord, Slack, and iMessage. On TryOpenClaw.ai, Telegram setup takes under 60 seconds. WhatsApp too. You don't touch BotFather or the Meta developer portal.

The --dangerously-skip-permissions problem

Claude Code has a permission system. Every tool call can trigger a prompt asking you to approve it. That's fine when you're at your keyboard coding. It doesn't work for an unattended messaging bot.

From the docs: "If Claude hits a permission prompt while you're away from the terminal, the session pauses until you approve locally." Your bot just stops responding. Someone messages at 2 AM and gets silence until you wake up and click "allow."

The official solution is --dangerously-skip-permissions. Anthropic named that flag with full intention. It bypasses all safety prompts. For a code editor running on your laptop, that's a calculated risk. For an always-on bot that strangers can message? The flag name tells you everything about how Anthropic views that use case.

OpenClaw's permission model was designed for autonomous operation from day one. Skill-level access controls, per-user permissions, and admin oversight that don't require disabling the safety system to function.

Pricing doesn't scale for messaging

Claude Code requires a Claude Pro ($20/month) or Max ($100-200/month) subscription. Usage counts against your allocation. Every message your bot processes burns through that quota.

A moderately active Telegram bot might handle 50-100 conversations a day. Each conversation involves tool calls, context processing, and response generation. On a Pro plan, you'd hit usage limits fast. On Max, you're paying $100-200/month for a Telegram bot that still requires you to maintain a server. Plus the server costs.

TryOpenClaw.ai is one flat monthly price — $39/month — with AI included. No per-message charges. No quota anxiety. No surprise bills.

Research preview means research preview

Channels require Claude Code v2.1.80+. They only work with claude.ai login — no API key auth. Console authentication isn't supported. Team and Enterprise orgs need admin opt-in. The --channels flag syntax "may change based on feedback." Only Anthropic-allowlisted plugins work during the preview.

None of this is criticism. New features start somewhere. But if you're choosing a foundation for your business's messaging agent today, "research preview with everything subject to change" is a different proposition than software that's been handling production messaging workloads for months.

Different tools for different jobs

Claude Code is an excellent coding assistant. Genuinely impressive. The channels feature makes sense for developers who want CI alerts or monitoring events pushed into their coding session. That's a real use case and it's well-designed for it.

But a code editor that can receive Telegram messages is not a messaging agent. An agent needs always-on availability, persistent memory, multi-platform support, proactive scheduling, business integrations, and a permission model that works without human supervision. OpenClaw has all of that. Claude Code channels have none of it.

The gap isn't quality. It's category. A Ferrari is a better car than a fishing boat. But you wouldn't take it offshore.

On TryOpenClaw.ai, you sign up, pick your messaging app, and you're chatting with your agent in under 60 seconds. Always on. Persistent memory. WhatsApp, Telegram, Discord, Slack, iMessage. No terminal to babysit. No flags to set. No permissions to bypass. Managed hosting starts at $39/month.

Frequently asked questions

What are Claude Code channels?

A research preview feature in Claude Code (v2.1.80+) that pushes messages from Telegram or Discord into your running CLI session via MCP servers. Claude reads the message and can reply back through the same channel. The session must be open for messages to arrive.

Can Claude Code channels replace OpenClaw?

No. Claude Code channels lack persistent conversation memory, proactive scheduling, WhatsApp support, always-on availability without server management, and a permission model designed for autonomous operation. They're a notification pipe for developers, not a business messaging agent.

Is Claude Code channels free?

You need a Claude Pro ($20/month) or Max ($100-200/month) subscription. Usage counts against your plan's allocation. Server costs for always-on operation are extra.

Does Claude Code channels support WhatsApp?

No. Only Telegram and Discord during the research preview. You could build a custom WhatsApp channel, but you'd need to handle the Meta Business API, webhook verification, and messaging policies yourself.