Send a task from your phone on the commute, Claude processes it on your computer, and you come back to finished work. No dedicated Mac Mini, no Node.js, no 48-step startup sequence. Dispatch is a new Claude Cowork feature that does what people have been trying to build with OpenClaw, just without the pain. Here's how to set it up in 5 minutes and what I use it for.
Key takeaways:
- Dispatch = a persistent conversation with Claude running on your computer, controlled from your phone via the Claude mobile app
- Claude works in a sandbox, files stay local, you approve what it can access
- Setup: download Claude Desktop + mobile app, pair them. Done in 5 minutes
- OpenClaw requires a dedicated server, Node.js 22+, a WebSocket gateway, and extensive configuration. Dispatch needs none of that
- Available for Max subscribers ($100/month), Pro ($20/month) coming in the next few days
What is Dispatch and why I'm writing about it
Three weeks ago I wrote about Claude Cowork and my daily briefing. Cowork scans my email, calendar, Sentry every morning and prepares a structured overview. Works great, but with one condition: you have to be sitting at your computer.
Dispatch changes that. It's a new feature in the Cowork tab of Claude Desktop (research preview), announced by Felix Rieseberg from Anthropic:
We're shipping a new feature in Claude Cowork as a research preview that I'm excited about: Dispatch!
— Felix Rieseberg (@felixrieseberg) March 17, 2026
One persistent conversation with Claude that runs on your computer. Message it from your phone. Come back to finished work.
To try it out, download Claude Desktop, then pair… pic.twitter.com/r6OH46Ll89
The concept: Claude works in a sandbox on your machine, files stay local, and you approve what it can access from your phone.
An expensive Mac Mini and OpenClaw: a solution looking for a problem
Earlier this year, there was a wave of excitement around OpenClaw. Czech productivity author Petr Ludwig bought a Mac Mini, named his AI agent Lana, set up OpenClaw, and posted on Facebook about running 6 AI agents that assign tasks to each other. Lana "read" his two books and listened to 200 podcast episodes overnight.
Sounds impressive. It does require a certain faith that a bot won't decide to overthrow humanity after listening to 200 podcasts. And also:
- A Mac Mini M4 (hundreds to over a thousand dollars) as dedicated hardware that computes nothing, just calls cloud APIs
- Node.js 22+ (24 recommended)
- Installation via npm or curl script that regularly fails
- WebSocket gateway on port 18789
- A daemon process you need to maintain
- A 48-step, 9-phase startup sequence (not exaggerating, it's in the code)
- Messaging bridge configuration (WhatsApp, Telegram...)
And here's the thing: OpenClaw runs locally, but the AI models themselves are called in the cloud. You're sending requests to Claude API, GPT API, Gemini API. That Mac Mini is essentially an expensive proxy server. Two years of Claude Max subscription (~$2,400) cost less than the hardware while giving you access to superior models.
The community on GitHub acknowledges this themselves. There's an open issue requesting a "non-tech friendly Setup Wizard" because the current setup is "built by developers, for developers." New users regularly get stuck.
You don't need a dedicated server to tell AI what to do from your phone. You need an app you can download in a minute.
How to set up Dispatch: step by step
The entire setup takes 5 minutes. No terminal, no npm, no configuration needed.
- Download Claude Desktop from claude.com/download (macOS or Windows). Install and sign in.
- Download the Claude mobile app from the App Store (iOS) or Google Play (Android).
- Pair your phone with your computer in the mobile app. Sign in with the same Claude account and the app will offer to pair with your desktop.
- Open the Cowork tab in Claude Desktop. Send a task.
- Switch to your phone. You can see progress, get updates, refine the task, or approve actions.
Done. The entire 48-step startup, WebSocket gateway, and daemon from OpenClaw replaced by two downloads and a pairing.
Important: Claude Desktop must stay running on your computer. Dispatch doesn't run in the cloud. Claude works locally in your sandbox, the phone acts as a remote control. If you shut down your computer or close Desktop, the session stops.
What I use it for in practice
Dispatch isn't just "phone as terminal." It's a way to assign work the moment an idea hits, without losing it.
- Dashboard reports: "Scan Sentry for the last 24 hours, sort issues by impact, prepare a report." I send it from my phone on the way to work, a finished overview waits on my monitor
- Meeting prep: "Go through emails from client X this past week, summarize open items, prepare an agenda." Sent from my phone 20 minutes before a call
- Data processing: "This folder has a CSV with Q1 orders. Analyze trends, create charts, summarize findings." Claude processes local files while I monitor progress on my phone
- Flight hunting: Felix himself mentions Claude finding him a better seat on a flight. Dispatch has access to the browser on your computer, so it can search and compare offers
- Content repurposing: "Take the latest blog post, prepare a LinkedIn post, X thread, and newsletter draft." I assign it in the evening, finished drafts wait in the morning
The key is that Claude has access to everything running on your computer: files, browser, installed tools. And you control it all from your pocket.
I need help setting up AI workflows
Dispatch vs. OpenClaw: comparison
| Claude Cowork Dispatch | OpenClaw | |
|---|---|---|
| Setup | Download 2 apps, pair | npm install + daemon + gateway + messaging bridge config |
| Time to start | 5 minutes | 30-120 minutes (if everything works the first time) |
| Dedicated HW | No (runs on your computer) | Recommended (Mac Mini, server) |
| Control via | Claude mobile app, claude.ai | 21+ channels (WhatsApp, Telegram, Slack...) |
| AI models | Claude (Anthropic) | Multi-provider (Claude, GPT, Gemini...) |
| Sandbox | Native, files stay local | Depends on configuration |
| Action approval | Yes (from phone) | Via Approval Manager |
| Browser access | Yes (native) | Yes (CDP) |
| Voice control | No | Yes (Voice Wake, TTS) |
| Price | $20-200/month (by plan) | Free (self-hosted) + LLM costs + HW |
| Target user | Anyone | Developers and power users |
Where OpenClaw wins: if you need to send messages via WhatsApp or Telegram, want to use multiple AI models, or need voice control. OpenClaw is a more universal platform.
Where Dispatch wins: everything else. Simpler setup, no infrastructure, native sandbox, phone-based approvals. And most importantly: you don't need to be a developer to get it running.
Limitations and what to expect
- Research preview. Dispatch is an early version. Felix himself says more is coming in the next days and weeks
- Computer must stay on. Claude works locally. If you close your laptop, the session stops. For always-on mode, you need a desktop that doesn't sleep (but not an expensive Mac Mini)
- Max subscribers first. Dispatch is currently available for Max plan ($100/month). Pro ($20/month) is coming in the next few days
- Claude mobile app only. No WhatsApp, Telegram, or other messaging channels yet. If you need that, OpenClaw is (for now) the only option
- Quota. Cowork tasks consume more tokens than regular chat. On Pro, you can handle a few tasks per day. For heavy use, you'll need Max
Bottom line: you don't need a server, you need an app
OpenClaw showed that an AI assistant can run on your machine and be accessible from anywhere. But the setup requires technical skills most people don't have, and buying dedicated hardware that just calls cloud APIs doesn't make financial sense.
Dispatch solves the same thing with two apps and one pairing. If the Claude mobile app is a sufficient control channel for you (and for most people it is), this is clearly the simpler path.
If you're already using Cowork (like for a daily briefing), Dispatch is the natural next step. Instead of sitting at your computer, you send tasks from anywhere.
I need help with AI automation
