Run parallel agents

Create one workspace per independent task. Give each agent a concrete goal, relevant context and a clear definition of completion.

Copperbar supports installed Claude Code, Codex, Cursor and OpenCode CLIs. Sign in with the provider’s own tooling before starting a session. Model access, usage limits and billing remain with the provider.

Keep responsibilities clear

Ask one workspace to fix the failing test and another to investigate a separate issue. Avoid having multiple agents modify the same branch or rely on each other’s uncommitted files.

An agent reporting success is a useful signal, not proof that a task is finished. Review the diff and run checks before accepting its work.

Plan dependent tasks

Open More → Task plans to build a project-level workflow. Add each task's instructions, agent setup and dependencies, then save the plan. Saving does not start an agent. Choose Start plan when you are ready to use your provider account.

Commit the project changes you want included before starting. A run keeps its original plan revision and base commit, even if you edit the saved plan later. Each task gets a separate workspace. A dependent task can inherit the saved code snapshot from one completed dependency; other dependencies contribute their recorded output. Copperbar does not automatically merge several branches.

Set the parallel-task limit and the time limit per task. Ownership paths reserve overlapping areas of the project while tasks run; they do not restrict which files an agent can access. Leaving ownership blank reserves the whole project. Review is a workflow role, not a read-only permission mode.

Run details show preparation, execution, saved snapshots, errors and blocked tasks. Completed means the agent finished and Copperbar saved its snapshot; it is not an independent approval to merge. Open the task workspace to inspect the changes and run checks.

After a failure or interruption, select the unsuccessful tasks you want to retry. Their blocked dependants are included, and successful snapshots are reused without running those agents again. Restarting the app does not replay interrupted work. Cancel a run to stop new task admissions and its owned sessions; if cleanup needs attention, use Retry stopping tasks before starting replacement work.

Drafting, reviewing and exporting plans remain available without Pro. New runs and retries are Pro features; preview builds may include them. An already started, unchanged plan can finish after trial expiry. Provider usage is billed separately by your agent provider. Run exports contain private instructions and agent output; share them deliberately.

Stop and resume

Use the session controls to stop active work. Terminal sessions and agent processes depend on the host machine. Save changes and checkpoint important progress before shutting down or replacing the application.