Skip to content

Your First Agent

In Fletch, each agent works in its own workspace — an isolated checkout of your repo (a lightweight clone that borrows your repo’s history), inside a macOS sandbox. Two agents can never edit the same files, because they’re working different checkouts. Spawning an agent is how you create a workspace.

  1. Click + New agent on a project (or press ⌘N) to open a draft.
  2. Under “What should be the first task?”, describe the task in the composer.
  3. Pick the agent and model from the model picker (Claude Code is the default). Optionally change the base branch (defaults to main) or reroll the generated name.
  4. Press Enter to spawn.

Fletch creates an isolated checkout for the agent — forked from the freshest state of your base branch — and launches the agent’s CLI there under a sandbox. The draft pane shows exactly where it will live before you spawn.

  • The chat view shows the agent’s reasoning and tool calls as a readable transcript.
  • If you enable the experimental native view, you can flip to the agent’s raw terminal (available after its first turn).
  • Turn on the Code, Git, Run, and Terminal side panels in Settings to review diffs, manage git, run your project, and open a shell in the workspace.

Each agent runs under a per-agent sandbox — macOS Seatbelt by default, or an opt-in Docker container — that denies writes everywhere except its own workspace, temp directories, and the agent’s own config/cache. Under Seatbelt it can still read files and use the network. How the two engines differ, and why the checkout design matters, is covered in Security & Sandboxing.

When you’re done, archive the workspace to hide it while keeping a restorable snapshot, or discard it to remove the checkout and its branch permanently. Both are one click in the workspace. And if a session is worth continuing in a new direction instead, you can fork it.