A question, in context
Collaboration does not have to leave the work.
People and agents need a place to ask for a check, answer it, and keep moving. Those exchanges are most useful when they remain connected to the work that prompted them.
Durable collaboration for software teams
Software engineering is collaborative. AI should be, too.
Developers share context, ask questions, investigate problems, review work, leave notes, and hand work to one another. Coding agents should participate in that work—not disappear into isolated conversations.
Dispatch is a local-first shared workspace where humans and coding agents collaborate using persistent terminals, worktrees, shared context, and reusable capabilities.
A working environment
The useful parts of working together can be visible.
These are small moments in a piece of engineering work: seeing what is underway, keeping the useful context close, and getting another set of eyes on the same change.
Context that remains usable
The starting point should be there when someone joins.
Links, files, and decisions need to be easy to find at the moment they become relevant. Keeping them with the work avoids making every handoff a reconstruction.
A reviewer is working
Review is a teammate activity, not a terminal state.
A reviewer can join an agent’s work while it is still unfolding. The person directing the work can see that a second perspective is active without turning review into a separate workflow.
The reviewer found something
Feedback should stay close enough to be acted on.
Review is useful when concerns retain their location and status. The developer and agent can work from the same concrete record rather than translate a verdict back into tasks.
The shared workspace
Humans and agents work with the same durable objects.
Dispatch is the space around an agent session where people can see what is happening, contribute context, and return to the work. The browser is the control surface; the work continues in the local environment that owns it.
Persistent terminals
Agents keep working in terminals that survive a browser session, a disconnect, and the time between decisions.
Independent worktrees
Humans and agents can change the same repository without borrowing or disturbing each other’s directory.
Shared context
Messages, screenshots, media, pins, whiteboards, and Brain preserve the evidence behind a decision.
Reusable practices
Review personas, jobs, notifications, and repo-aware tools make a good way of working available again.
A realistic handoff
Leave at six. Pick up the same investigation at nine.
- A developer launches an agent to investigate a bug in an isolated worktree.
- The agent finds a staging URL, shares a screenshot, and pins both where the team can find them.
- A reviewer launches a persona, reads the same artifacts, and leaves feedback in the shared workspace.
- Later, the developer joins from another device. The terminal is still running; the worktree, conversation, evidence, and feedback are all still available.
Local-first
The browser is a control surface over your environment.
Dispatch runs against the repositories, credentials, coding-agent CLIs, and compute already available on your macOS or Linux machine. It does not replace that environment; it makes its ongoing work easier to see, share, and direct from a browser.
repositoriescredentialsagent CLIslocal compute
Why capabilities?
Collaboration needs room for the work to change shape.
Capabilities are useful independently: start an isolated worktree, share a screenshot, leave a message, ask for a review, store a decision in Brain, or schedule recurring work. They also compose naturally when an investigation takes an unexpected turn.
That is why Dispatch is built from capabilities rather than a rigid workflow engine. Workflows can be repeatable without becoming the only way collaboration is allowed to happen.
An ongoing experiment
Useful context should not disappear between sessions.
Dispatch grew out of using coding agents on real software projects and watching useful context get trapped in terminal sessions, browser tabs, temporary screenshots, and isolated conversations.
It is an ongoing experiment in making collaboration with AI feel more like collaboration with another engineer: durable, visible, and grounded in shared work. It is not an attempt to replace developers; it is an attempt to give them a better place to work alongside agents.
Continue reading
See how the workspace is built.
Read the principles behind the experiment, see why its pieces exist in the architecture, or run Dispatch locally with your own team.