Claude Code orchestration: from one terminal to parallel agents

Claude Code is Anthropic's agentic CLI: it reads your codebase, edits files, runs commands, and works a task to completion in your terminal. If you're reading this, you almost certainly use it already, so the question isn't what it is, it's how to run more than one of it.

Orchestration is the layer around the sessions: giving each one an isolated branch so they can't overwrite each other, a scoped task so they don't drift, a status view so you know what's running, and a review path so finished work becomes a PR instead of a pile of uncommitted changes. It's the difference between a CLI and an agentic IDE workflow.

None of this is a flaw in Claude Code, the CLI is deliberately scoped to one session in one working directory. The friction only appears when you try to run several at once:

There are three honest approaches, and each is right for someone. Terminal multiplexing is free and works everywhere. Manual worktrees give you real isolation with some git bookkeeping, we cover the full recipe in parallel agents with git worktrees. An orchestrator automates the whole loop.

AIDEN is a desktop workspace that runs on top of your local Claude Code and/or Codex CLI, it handles the workflow while the CLIs handle the model calls. The board has five columns: Stories → Spec Review → In Progress → Review → Done. A full comparison with using the bare CLI lives at AIDEN vs Claude Code.

1. 1

   Write a story

   Describe what should ship in plain language on the board. AIDEN has already analyzed your codebase, so the agent gets project context for free.
2. 2

   Approve the spec

   AIDEN drafts a spec, goal, acceptance criteria, technical approach. Since v1.5.21 this approval is an enforced gate: no agent starts until you sign off.
3. 3

   Agent implements on its own branch

   Launching the story creates an isolated git branch/worktree and starts a Claude Code session there with the approved spec as context. Parallel stories can't touch each other's files.
4. 4

   Agent iterates; you watch status, not keystrokes

   The agent can run your tests and iterate on failures (best-effort, see the note below). The board card shows where each story stands.
5. 5

   Review the diff, ship the PR

   When a story reaches Review, you read the diff in the built-in git panel, optionally run the LLM PR review pass, and open a pull request in one click.

If Claude Code already works in your terminal, setup takes a few minutes. AIDEN uses your existing installation and credentials, keys stay in ~/.claude, and your code never leaves your machine.

1. 1

   Verify your Claude Code install

   Check the CLI is on your PATH; install it first if not. AIDEN requires at least one of Claude Code or Codex CLI installed locally.

   claude --version
   # not found? install it:
   npm install -g @anthropic-ai/claude-code

2. 2

   Install AIDEN

   Download the macOS 12+ app from aidenapp.org (Apple Silicon and Intel, signed and notarized) and drag it to Applications. Free for 1 project, no credit card.
3. 3

   Create a project

   Point AIDEN at a local git repo. It analyzes the codebase and builds a technical and business overview that every agent session receives as context.
4. 4

   Keep your MCP setup

   AIDEN inherits your existing Claude Code MCP configuration automatically, no re-wiring. If you haven't set up MCP yet, see MCP servers for AI coding.
5. 5

   Run your first parallel stories

   Create two stories, approve both specs, and launch them. Each gets its own branch and session; review each diff as it reaches the Review column.