Dispatch
Menu
How it worksIdeasArchitectureDocs Principles
GitHub ↗

How it works

A shared workspace, not another agent chat.

Dispatch gives humans and coding agents the same place to work with persistent terminals, isolated changes, shared artifacts, and reusable collaboration practices.

Software work already has a shared surface

A real engineering task is more than a prompt and a response. It has a repository, a terminal, a branch, a staging URL, screenshots, questions, review feedback, and decisions someone will need to find again. Dispatch brings that surface into view instead of leaving it scattered across tabs and one-off conversations.

Human asks, guides, decides
Coding agent investigates, changes, reports
Reviewer inspects, questions, approves
Future teammate returns, understands, continues
Dispatch shared workspace
terminals · worktrees · messages · media · pins · whiteboards · Brain · personas · jobs
Everyone meets the same durable work and context. They do not need to reconstruct it from a private chat history.

What persists

Execution: a terminal and the agent inside it keep running in tmux when a browser tab closes. Changes: a worktree and branch remain separate from the work already in progress. Context: messages, screenshots, media, pins, whiteboards, and Brain keep the useful record close at hand. Practice: personas and jobs preserve a review style or recurring task without reducing every situation to a predefined pipeline.

A bug investigation, without losing the thread

  1. A developer launches an agent to investigate an intermittent production bug. Dispatch creates an isolated worktree so the investigation does not interrupt anyone’s current changes.
  2. The agent runs the app, watches logs, and discovers a staging URL. It pins the URL and shares a screenshot of the failure rather than burying both in terminal output.
  3. The developer adds a message with a hunch. A review persona opens the diff, sees the same screenshot and pinned URL, and leaves structured feedback.
  4. The developer closes the browser. The agent continues in its terminal, the artifacts remain shared, and the state is not held hostage by a live chat window.
  5. That evening, the developer returns from another device. They can see the terminal, the worktree, the feedback, the staging link, and the decision trail—then decide what should happen next.

Why this is not a workflow engine

The point is not to make every task follow a fixed sequence. An investigation can become a design conversation. A review can reveal a new bug. A human can decide the next step is to wait, split the work, inspect production, or stop. Dispatch makes the underlying collaboration capabilities available in each of those moments.

Those capabilities can support a repeatable workflow when one is useful. But the workflow sits on top of a shared workspace, rather than replacing it.