---
title: "Claude Dispatch: what is public, and what we have not tested"
description: "Claude Dispatch lets you message Claude from your phone while it works on your desktop. We have not tested it: what is public, and who said each part."
url: https://personalagentbench.com/claude-dispatch/
last_updated: 2026-11-05
---

# Claude Dispatch, the phone-to-desktop thread inside Cowork

Claude Dispatch is a feature of Anthropic's Claude Cowork that lets you message Claude from your phone while it carries out the task on your desktop computer, using the files, connectors, plugins and apps on that machine. Anthropic launched it on 17 March 2026 as a research preview. As of 5 November 2026, Anthropic's help page still says it "isn't available to new users." We have not tested Dispatch and we have not taken it apart. It is on [the landscape list](https://personalagentbench.com/landscape/) as recorded and nothing more, so this page is an account of what Anthropic and a small number of reviewers have published, who said each part, and what a teardown would have to settle.

- Dispatch is not a separate app. It is one continuous conversation inside Cowork that you can reach from the Claude mobile app or the Claude Desktop app.
- The work happens on your own computer, which has to stay awake with the desktop app open. Anthropic now points people without Dispatch to Cowork in the cloud instead.
- It is closed to new users. Anthropic's pages say existing users can keep using it "for now" and give no further date.
- Nothing here is a measurement. Dispatch is not in our exam, has no score and no teardown, and the hands-on findings below belong to the reviewers who made them.

## What Claude Dispatch is, in Anthropic's words

The main source is Anthropic's help article, now titled [Assign tasks from anywhere in Claude Cowork](https://support.claude.com/en/articles/13947068-assign-tasks-from-anywhere-in-claude-cowork). It says that with Dispatch "you can message Claude from your phone and have it work on your desktop computer," using local files, connectors, plugins and apps, "then come back to the finished work." The examples it gives are office chores: pull data from a local spreadsheet into a summary, search Slack and email to draft a briefing, build a presentation from Google Drive files, tidy a folder.

Anthropic's [release notes](https://support.claude.com/en/articles/12138966-release-notes) date the launch to 17 March 2026 and describe it as a "persistent agent thread" for managing tasks in Cowork, rolled out to Max plans that day and to Pro plans over the following two days. Six days later, on 23 March, Anthropic's blog post [Put Claude to work on your computer](https://claude.com/blog/dispatch-and-computer-use) described Dispatch as "a new feature in Claude Cowork (and now available in Claude Code)" and paired it with a new computer use preview, so Claude could operate desktop apps on your behalf while you were away.

Our landscape entry is for this delegation feature, not for Claude's chat product in general, the same line we drew for [OpenAI Dots](https://personalagentbench.com/openai-dots/) inside ChatGPT.

## How Claude Dispatch works: one thread, two kinds of session

The help article describes a single persistent thread that "doesn't reset." Claude keeps context from earlier tasks, so a message sent from a phone on the way to work and a follow-up typed at the desk land in the same conversation.

When you hand it a task, the article says Claude decides what kind of work it is and starts a session to match: development tasks go to Claude Code, knowledge work goes to Cowork, and each appears in its own sidebar. Claude then reports back with the outcome, which the article lists as things like a spreadsheet, a memo, a comparison table or a pull request, "rather than showing you every step." You get a push notification on your phone when a task finishes or when Claude needs your go-ahead.

Three further pieces sit on top of the thread, all from the same article:

- **Memory.** Claude carries your preferences and projects across sessions, and the article says you can view, edit and delete what it remembers at any time.
- **Scheduled tasks.** Recurring work can run on a schedule. [Anthropic's Cowork documentation](https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile) says new scheduled tasks on Pro, Max and Ultra run in the cloud after the 6 October change; existing local schedules remain on the computer. Cloud tasks need the desktop app open when they use local files.
- **Computer use.** Claude can operate apps on the desktop, for example updating a spreadsheet in Excel or navigating an internal dashboard. The article says computer use is not part of the Linux beta.

The thread is also a constraint. Under current limitations the article says there is "no way to start a new thread or manage multiple threads." Everything you ever ask Dispatch lives in one conversation.

## Claude Dispatch availability: limited beta, closed to new users

This has changed most since launch, and launch-week coverage does not reflect it.

At launch, Dispatch was for paying subscribers. The release notes say Max first, then Pro. John Voorhees of MacStories wrote on launch day that only Max subscribers had it and that Anthropic had promised Pro users access within "a few days."

The current help article says the capability "is in limited beta on some Pro, Max, and Ultra plans." It adds a note at the top: "Dispatch isn't available to new users. If you already use Dispatch, you can keep using it for now." If you do not see Dispatch in the Cowork side panel, the article tells you to use Cowork in the cloud instead.

Dispatch is not the same thing as [Claude Duet](https://personalagentbench.com/claude-duet/), the separate proactive agent app Anthropic opened to a US waitlist on 21 October 2026. Duet runs in Anthropic's cloud with no access to your computer; Dispatch exists to work on your computer.

The surrounding changes explain where Anthropic is pointing people. On 16 September 2026, Anthropic's blog announced that [Cowork and chat are merging into one Claude](https://claude.com/blog/cowork-is-now-claude), rolling out to Pro and Max first and now covering Ultra, with Team plans in beta and Enterprise plans where enabled. The help article on [the merged experience](https://support.claude.com/en/articles/16761823) lists "Dispatch isn't available to new users" among its current limitations, and separately says you can start a task from your phone "and Claude keeps working in the cloud." Anthropic's page on [Cowork across web, desktop and mobile](https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile) says that since 6 October 2026, new Cowork tasks on Pro, Max and Ultra run in the cloud, and the "Only on your computer" setting was removed. For new tasks that must run on your computer, the same page now points to Claude Code in the desktop app.

What none of these pages says is whether Dispatch will be retired, or when. "For now" is Anthropic's phrase, and we are not going to read more into it than that.

## What Cowork Dispatch costs

We found no separate price for Dispatch. It comes with the plan. Anthropic's [pricing page](https://claude.com/pricing) lists Pro at $20 a month billed monthly, or $17 a month on an annual subscription billed up front, Max from $100 a month and Ultra at $200 a month. The help article lists a Pro, Max or Ultra plan among the requirements, alongside the latest desktop and mobile apps and an internet connection on both devices.

The merged-experience help article says that "everything you do with Claude counts toward your plan's usage limits," and that longer agentic tasks generally use more than a quick question. It does not give a figure for a Dispatch task. The one first-hand account of cost we found is Ben Patterson's [PCWorld column](https://www.pcworld.com/article/3122909/claude-dispatch-is-the-future-brace-for-the-quota-shock.html) of 24 April 2026: on the $20 Pro plan, one Dispatch exchange that triggered a scheduled Cowork automation used 66 percent of his five-hour usage window "in a matter of minutes." He says it is not clear what consumed it. That is one session on one account, not a rate.

The practical point is that paying for a plan no longer gets you Dispatch. A new Pro, Max or Ultra subscriber is, by Anthropic's own note, a new user.

## Setting up Cowork Dispatch, and the computer that has to stay awake

The help article's steps are short. Update the desktop and mobile apps, open Cowork, click "Dispatch" in the left side panel, click "Get started," then use two toggles on the next screen: one gives Claude access to your files, the other keeps your computer awake. Then "Finish setup," and the conversation syncs across both devices. The desktop app runs on macOS, Windows x64 or Linux, per the requirements list.

Voorhees described a slightly different flow at launch: after updating the Mac app, Dispatch prompted him to scan a QR code to pair his iPhone, and a Dispatch entry then appeared in the iOS app's sidebar.

Amanda Caswell's [Tom's Guide piece](https://www.tomsguide.com/ai/i-sent-claude-a-task-from-my-phone-and-it-finished-it-on-my-laptop-without-me-touching-a-thing), published on 24 March 2026, says "there is no 'Dispatch' button," which conflicts with the help article and MacStories. We cannot tell whether she was describing a different build or simplifying.

Dispatch work runs on your desktop, so the help article says "your computer must be awake and the app must be open for Claude to work on tasks." A cloud session, by contrast, "runs on Anthropic's servers and keeps working even when your computer is off" (both read 5 November 2026). Anthropic's [help article](https://support.claude.com/en/articles/13947068-assign-tasks-from-anywhere-in-claude-cowork) and [walkthrough](https://academy.claude.com/use-cases/handle-a-request-while-away-from-your-keyboard) describe the keep-awake setting. Connected services such as email live online rather than on your disk, so before you leave a task running while you are away, check which of the two routes it is on.

## What Claude Dispatch can reach, and Anthropic's own warning

Anthropic's help article is unusually direct about the risk. It says Dispatch gives you access, from your phone, to "everything on your desktop through Claude," including files, connectors, plugins and apps through computer use. It describes the setup as a chain in which phone instructions can trigger "reading, moving, or deleting local files," using connected services, and "controlling your browser." It warns that a manipulated instruction or a phishing link "could cascade into actions that are difficult or impossible to undo," and ends with: "Only connect these agents if you're comfortable with what they could do, not just what you intend them to do."

The [computer use help page](https://support.claude.com/en/articles/14128542-computer-use-safety) adds detail. Claude asks permission before accessing each application, some apps are blocked by default, and you can keep a blocklist. Claude takes screenshots to navigate, so it can see whatever is on screen. The page says Claude is trained to avoid risky operations such as transferring funds or deleting files, and then says "these safeguards aren't perfect." It also states that computer use "has no sandbox between Claude and your applications."

On approvals, the Academy walkthrough recommends setting Slack and Gmail send access to "needs-approval," so Claude shows a drafted message before it goes out, and says computer use is off by default in Dispatch settings. It also says files "stay local" when Claude reads them and that nothing is uploaded to a cloud service. That describes where the source document stays. The walkthrough does not describe what is sent to Anthropic's servers for the model to read it, and we did not find a page that does for Dispatch specifically.

All of this is Anthropic describing intended behaviour, not something we have observed.

## What reviewers found when they used Claude Dispatch

The three hands-on accounts we could read in full date from March and are not structured tests. Later reports exist; this section describes the evidence we used, before the September and October changes.

[John Voorhees at MacStories](https://www.macstories.net/stories/hands-on-with-claude-dispatch-for-cowork/), writing on 17 March 2026, paired an iPhone with a Mac Studio and published a table of thirteen attempts. By his own marks, six worked, one partly worked and six failed. Finding a screenshot by its contents, summarizing Notion notes, adding a URL to Notion and summarizing his latest email all worked. Opening the Shortcuts app, sending a screenshot via iMessage, listing Todoist tasks (an authorization error), listing Terminal sessions and reading the active Safari tab all failed. He called it "rough around the edges but shows promise," said it "is currently slow," and concluded it was "about a 50/50 shot whether what you try will work." His verdict on using it away from a desk: "That's not good enough to rely on."

[Amanda Caswell at Tom's Guide](https://www.tomsguide.com/ai/i-sent-claude-a-task-from-my-phone-and-it-finished-it-on-my-laptop-without-me-touching-a-thing) tested Dispatch together with computer use a week later. She reports sending a task from her phone and finding it done when she got back to her laptop, but publishes no task-by-task results. She advises checking the output because Claude "can still make mistakes," and writes that she is "still on the fence about how often I will use this feature," because reviewing the work afterwards cuts into the time saved.

[Ethan Mollick](https://www.oneusefulthing.org/p/claude-dispatch-and-the-power-of), writing in One Useful Thing on 31 March 2026, describes two tasks sent from his phone: a morning briefing drawn from his calendars, email and online channels, and checking whether a chart on slide 3 of a presentation was current, then updating it. He says it "got slightly stuck" when a site blocked a download, called the results "very impressive," and called computer use "impressive as a concept and error-prone in practice."

Jenny Ouyang's [18-task review on Build to Launch](https://buildtolaunch.substack.com/p/claude-dispatch-use-case-breakdown-demos-vs-reality) (8 April 2026) is a fourth, longer account. Its free section gives one figure, computer use "about 60% reliable" on her hardware; the task-by-task results are behind a paywall, so we have not used them. We kept to named reviewers who show what they tried. A thirteen-row table is evidence about one Mac in one week, not a rate.

## Claude Dispatch troubleshooting and open questions

Existing users should check app updates, connectivity, desktop wakefulness and file permissions. If Dispatch is absent, the official help article directs new users to cloud Cowork, which keeps working with the computer off but reaches local folders only while the desktop app is open.

If Dispatch ever went through [the process we use](https://personalagentbench.com/how-we-take-them-apart/) for our [teardowns](https://personalagentbench.com/agents/), these are the questions it would need to settle.

Where the work runs when a task is routed to a cloud session versus the local desktop, now that Anthropic has moved new Cowork tasks to the cloud while keeping Dispatch tied to an awake machine. What the agent actually holds at a given moment: the connectors, plugins, folders and apps it can reach, as distinct from the ones a user meant to grant. What the approval gate is attached to in practice, given that the help pages describe per-app permission for computer use and per-connector "needs-approval" for sends, and whether a message from a stranger arriving in a connected inbox can steer it. What memory holds in a single thread that never resets, and whether deleting one remembered item removes it everywhere it was used. And, given Anthropic's own phrase "for now," whether an existing user's thread and memory survive if the feature is withdrawn.

Only observed behaviour goes on our [scorecard](https://personalagentbench.com/how-scoring-works/). Until then, Dispatch gets a page like the ones for [Poke](https://personalagentbench.com/poke-ai/), [OpenAI Dots](https://personalagentbench.com/openai-dots/) and [Grok Bot](https://personalagentbench.com/grok-bot/). A blank here is not a low score. It is a blank.

## FAQ

**What is Claude Dispatch?** A feature of Claude Cowork, from Anthropic, that gives you one persistent conversation you can message from your phone while Claude does the work on your desktop computer, using its local files, connectors, plugins and apps.

**Can I still sign up for Claude Dispatch?** According to Anthropic's help article, no. It says "Dispatch isn't available to new users" and that existing users can keep using it for now. It directs people without Dispatch to Cowork in the cloud.

**Does my computer need to be on?** Yes. Anthropic says the computer must be awake with the Claude Desktop app open while Claude works. Cloud sessions, by contrast, keep running when your computer is off.

**How much does Claude Dispatch cost?** There is no separate price. It required a Pro, Max or Ultra plan; Anthropic's pricing page lists Pro at $20 a month billed monthly, Max from $100 a month and Ultra at $200 a month. Usage counts toward the plan's limits.

**Is Claude Dispatch the same as Claude Code Dispatch?** The name needs context. Anthropic’s official Dispatch can route development tasks into Claude Code and knowledge work into Cowork, according to its [help article](https://support.claude.com/en/articles/13947068-assign-tasks-from-anywhere-in-claude-cowork). "Claude Code Dispatch" in [MindStudio's explainer](https://www.mindstudio.ai/blog/what-is-claude-code-dispatch) is a community relay that forwards Telegram messages to a local Claude Code session, which MindStudio says is "not an official Anthropic product." The [dispatch repository on GitHub](https://github.com/bassimeledath/dispatch) is a third, unrelated Claude Code skill that hands tasks to background workers.

**Have you tested Claude Dispatch?** No. It is not in our exam, it has no score, and we have not taken it apart. The hands-on findings here belong to the reviewers named above.

## Other agents we have not taken apart

- [**OpenAI Dots**OpenAI Dots are always-on agents inside ChatGPT. We have not tested one: what is public, what they cost, and who said each part.](https://personalagentbench.com/openai-dots/)
- [**Poke AI**Poke AI is an agent you text, made by The Interaction Company and now part of Cognition. We have not tested it: what is public, and who said it.](https://personalagentbench.com/poke-ai/)
- [**Cue by Manus**Cue by Manus gives each agent its own email, phone number and wallet. We have not tested it: what is public, and who said each part.](https://personalagentbench.com/cue-by-manus/)
- [**Copilot Tasks**Copilot Tasks is Microsoft's consumer Copilot agent, now moving into Cowork. We have not tested it: what is public, and who said each part.](https://personalagentbench.com/copilot-tasks/)
- [**Claude Duet**Claude Duet is Anthropic's proactive agent, in a US waitlist preview since 21 October 2026. We have not tested it: what is public, and who said each part.](https://personalagentbench.com/claude-duet/)
- [**nanoMuse**nanoMuse is an open-source personal agent in the shape of Meta Muse, on your own phone and computer. We have not tested it: what is public, and who said it.](https://personalagentbench.com/nanomuse/)
- [**Grok Bot**What can Grok Bot do? Its eight work roles, and how it works: one cloud computer your bots share, skills, routines, approvals. How it differs from Meta Muse.](https://personalagentbench.com/grok-bot/)

Every product we track, and how far we have got with each, is on [the full list](https://personalagentbench.com/landscape/).

---

Published by Personal Agent Bench (https://personalagentbench.com), run by the Zentor team. Zentor builds agent software of its own, so its products are not on the scorecard and will not be added.
