Claude Code just became a complete piece to build your own OpenClaw.

A week ago it was /loop for running skills on a schedule.

Two days ago it was /dispatch and cowork.

Now Claude Channels. Messaging from Telegram and Discord.

You message a bot from your phone, it pushes into a running Claude Code session. Claude processes it, reply goes back. All first-party.

I've been running OpenClaw for this exact use case. One AI assistant you reach through your existing messaging apps.

It worked, but it was hard to manage. I was constantly SSHing into my Mac Mini from Claude Code to fix my OpenClaw setup. And honestly, the main thing I ended up using it for was as a therapist.

There was also always a gray area around authentication. Third-party tools piping through Claude Code subscriptions. Maybe violating ToS, maybe not. Nobody was sure.

That gray area is gone now. Anthropic is saying "yes, inbound messaging to Claude Code is a supported pattern." VentureBeat already called it an "OpenClaw killer."

I set it up coupled hours ago. Both Telegram and Discord, the whole pairing and access control flow. It works. I want to give it a proper try and expand beyond the therapist use case.

What I want to explore next is per-channel personas. Different agents with isolated context, each serving a different channel through the same Discord bot.

Combine that with /loop and you get heartbeats. Claude checks something every N minutes, reports back through your messaging app.

Still a research preview. But I'm genuinely surprised by the speed of shipping. Anthropic just reached internal AGI or something like that.