Hand real work to an agent on your own machine
Open a task in your DoBox, hand it to an agent, and get it back for review. It runs on your own computer through Claude Code or Codex, on the AI subscription you already pay for, and works on documents and research as well as code.
Get startedDownload, install, pair, then send a task to an agent.
- Claude Code
- Codex
Hand a task to an agent the way you would hand it to a person
Nothing has to be automated first.
You start it, or a workflow does
Delegate by hand from the task in front of you, or let a workflow hand it off when a condition you set is met.
Give it a document, get a draft back
Hand it a report, an analysis or a piece of research. It needs no repository, and the draft comes back attached to the task for you to review.
Give it code, get a draft pull request
The agent works in its own git worktree, so the assignee’s branch and uncommitted work stay untouched. A person merges every time, by choice. The task tracks its checks and reviews live, with one click to fix failing checks or resolve comments.
Your machine, or our cloud
Our desktop companion runs on Macs with Apple Silicon today, and Intel Macs aren’t supported. Linux is next, then Windows, with no date for either yet.
-
this+that cloud
- Nothing to install
- For when nobody wants to set up a machine
-
Your Mac, with Claude Code or Codex
- No copy of the repository anywhere else
- On the subscription you already pay for
- Sends context to Anthropic or OpenAI, as in any terminal
-
Your Mac, with Qwen on Ollama
- No API call and no model bill
- The code, the model and the work stay on the laptop
- Quality depends on the model and your hardware
Claude Code on your Mac, from inside the app
A terminal drawer in the app talks to Claude Code on your Mac, on your own subscription. Permission requests and questions arrive as cards, and Move to DoBox turns the session into a tracked task.
- Pick which paired Mac runs the session
- Paste screenshots and documents for Claude Code to read
- Pick up a delegated run in the terminal and hand it back
A bug pipeline, end to end
Our own bug pipeline, as an example rather than a template. The inbox, the ownership map and the approval points are yours to decide.
-
1A bug report lands at support@
The AI tells it apart from a billing question
-
2It becomes a DoBox task
With the original email attached
-
3The Brain routes it
To the engineer you said owns that area
-
4The owner delegates it
To Claude Code or Codex on their own machine
-
5A draft pull request comes back
The task updates as each step happens
-
6The reporter gets a reply
A workflow answers once the review is approved
David KimAssigning work to an agent is mainstream now
GitHub Copilot’s coding agent takes an issue and opens a draft pull request. Linear lets you assign an issue to Devin, Codex or Cursor like a teammate. They’re good products, built on the bet we made early: describe the work, hand it to an agent, review what comes back.
GitHub Copilot and Linear
- Where the agent runs
- A cloud environment the vendor provisions, which makes setup easy
- Where the loop starts and ends
- An issue, then a pull request or a ticket update
this+that
- Where the agent runs
- Our cloud, or the assignee’s own checkout on the subscription they already pay for
- Where the loop starts and ends
- A message, then a reply to whoever sent it, with the tracker in the middle
When work starts inside the team, create the task in the DoBox and delegate it from there.
Questions about delegation
What do I need to run this today?
Does my code get sent anywhere?
What does it cost?
How is my data handled?
Delegate your first task
Knowledge work or code, either way. Free for 30 days, no credit card required. Then $35 per user per month, or $29 billed annually.
Get started