Worktrees

Parallelism in Alera is not “open more tabs.” Two agents sharing one working tree last about ten minutes before they trample each other. The isolation primitive is a Git worktree.

One worktree per task

When you create a linked workspace on a Git-backed project, Alera creates a real Git worktree:

• a new local branch from a source branch, or
• an existing local branch you already have

Each agent then sees only its own checkout. Diffs stay scoped. You can jump between contexts without stashing or remembering what was half-done.

Why this changes how you delegate

When handing a task to an agent costs one worktree instead of a negotiation with your current branch, smaller tasks get easier to start. The tooling nudges you toward the workflow that actually works with several CLIs at once.

From repo to running agents

1. Register a Git-backed project (local folder or clone).
2. Open a linked worktree workspace for each task you want in flight.
3. Launch Claude, Codex, Amp, fx, or any other CLI in that workspace’s terminals.

The remaining chore used to be setup: every new worktree needs its .env, its gitignored config, its install. That ritual now lives in alera.toml. When the work is done, pull requests and CI checks stay scoped to the same worktree so what you review is what the agent produced.

What is not a worktree

The primary workspace points at the project path itself. Use it for the default checkout. Linked workspaces are the extra checkouts.

Non-Git projects cannot create linked workspaces. They still get terminals, agents, and persistence; they just share one tree.

SSH worktrees (run the same model on a remote machine) are on the roadmap. They are not part of the current workbench.