Run agents in parallel

This guide shows you how to run multiple agents at the same time.

How parallel execution works

Each agent gets its own isolated worktree — a copy-on-write clone of the repo on its own branch. Agents never share a working directory, so they can work on the same repo simultaneously without conflicts.

Dispatch multiple agents

Via DMs

Open a DM with agent A and send a task
Open a DM with agent B and send a different task
Both agents dispatch simultaneously in their own worktrees

Via channels

In a group channel, @-mention multiple agents:

@alice refactor the auth module. @bob add tests for the payment service.

Both agents dispatch in parallel. Their output streams into the same channel, tagged with the sender.

Via tickets

Create multiple tickets
Assign each to a different agent
The TicketDispatcher dispatches all of them — each gets its own channel and worktree

Via pipelines

Use a pipeline with parallel branches:

Add multiple agent steps after the trigger
Don’t connect them to each other — they run in parallel
Add a join step to wait for all of them to complete

Monitor parallel agents

The dashboard shows all running agents across workspaces:

Agent name and workspace
Status (running, blocked, idle)
PID and start time
Current run summary

The agent status indicators change in real time: idle → running → idle.

Handle conflicts

If two agents modify the same file on different branches, you’ll see the conflict when you try to merge the second PR. Resolve it the same way you’d resolve any Git merge conflict.

To minimize conflicts:

Give each agent a distinct scope (different files, different modules)
Use the code graph to understand dependencies before dispatching
Review and merge agents’ PRs promptly

Resource considerations

Each running agent consumes:

A worktree (disk space, negligible with CoW)
A sandbox process (CPU and memory)
API tokens (costs accumulate per agent)

Monitor costs in Analytics or via the manage costs and budgets guide.