> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dureai.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Your first parallel workflow

> Run two provider CLIs, prove managed-session survival, and close the loop with a diff review.

This walkthrough demonstrates Dure's core loop: keep active agents visible,
keep managed work alive when the window goes away, and bring changes back for
human verification.

## 1. Start two different providers

Create two managed agent panes in the same project, choose different providers,
and give them separate names. Enable worktree isolation so each pane receives
its own branch and working directory.

Give the agents independent tasks. Spaces shows both providers, their activity,
and live diff counts without making you switch between repositories.

## 2. Arrange the work without restarting it

Drag a pane to another desktop or split group. The pane moves; its live provider
process does not restart. Use desktops for phases such as Launch, Review, and
Operate rather than treating every project as a separate application window.

## 3. Prove the survival boundary

While a managed local agent is producing output, quit Dure and open it again.
The pane should reattach to the same Hmux-owned process and restore its screen,
not launch a new provider conversation.

<Note>
  A provider started manually in an ordinary terminal is recognized, but that
  alone does not make it managed. Start from an agent pane or use an offered
  managed-conversion action before relying on app-restart survival.
</Note>

## 4. Review and respond

Open the agent's Diff pane. Review changes against the branch fork point,
including uncommitted and newly created files. Select a line, write a concrete
request, and send it back to the agent as terminal input.

## What you proved

* two native provider CLIs can work side by side
* Spaces remains the live map even when panes move
* a managed local session outlives the app window
* review feedback returns to the agent without leaving Dure

## Current boundary

App restart survival currently applies to managed local Hmux sessions. A host
crash or machine reboot is recovery, not uninterrupted survival; SSH agents do
not yet use the same Hmux path by default. Diff review is currently local-only.
Read [Current limits](/en/current-limits) before relying on these guarantees.

Next, learn the [Spaces model](/en/spaces-and-panes) or the exact
[session and recovery semantics](/en/session-model).