Show HN: Conductor, a Mac app that lets you run a bunch of Claude Codes at once

Original Article ↗ Hacker News Discussion ↗

Purpose and Workflow Model

  • Conductor lets users run multiple Claude Code agents in parallel by creating isolated git worktrees per session.
  • Supporters say this solves conflicts when multiple agents edit the same repo and makes better use of “agent lag” time by letting them switch between tasks.
  • Skeptics argue they can already do this with multiple terminals, tmux, and manual worktrees; for them, Conductor “gets in the way” compared to keyboard-only workflows.

Git, GitHub, and Local Repos

  • Initial behavior was to clone repos from GitHub with broad OAuth permissions and no obvious privacy policy, which drew strong criticism (full read-write on all repos, org settings, deploy keys).
  • The author responds that this was due to OAuth limitations and says they are moving to a GitHub App with fine-grained permissions and also supporting local GitHub CLI auth.
  • Several users want purely local git support with existing checkouts, no mandatory GitHub integration, and no repeated dependency installs; setup scripts (copying node_modules, env files, running installs) are suggested but seen as obscure.
  • Worktrees cause friction for untracked files (env files, submodules) and PR workflows; some consider the extra complexity not worth it for most AI coding tasks.

Features, UX, and “Feel”

  • Requests include:
    • Changing the default branch for new workspaces.
    • Custom “Open in…” commands (e.g., SourceTree).
    • Embedded terminal, “plan mode,” message queuing, and multi-repo workflows.
  • Some users feel Conductor loses the “feel” of native Claude Code: streaming output, escape-to-interrupt behavior, and tight terminal interaction.

Alternatives and Ecosystem

  • Multiple alternatives are mentioned: Crystal, Claude Squad, Plandex, container-use, par, vibe-tree, hydra, autowt, etc., many of them open source and focused on simple worktree or container management.
  • Some prefer minimal tools that “do one thing well” and work against existing local repos.

Broader Reflections

  • There’s debate about whether parallel agents truly increase productivity versus the human review bottleneck.
  • Concerns are raised about AI coding tools and data privacy in general; others respond that enterprise contracts, on-prem/cloud provider hosting, and not storing secrets locally mitigate risk.