Skip to content

runcastle vs Conductor

Both tools solve the same first problem: one Claude Code session at a time is a bottleneck, and running several by hand means juggling git worktrees in a pile of terminals. Conductor and runcastle both give every agent its own branch and its own checkout. What differs is what surrounds the agents — and which operating systems you can run them on.

The short version. Conductor is a polished native Mac app for running and reviewing parallel agent sessions — Claude Code, Codex, Cursor and OpenCode — each in its own git worktree, ending in a pull request. runcastle is a local web app for the work either side of that: it grills you into a spec, cuts the spec into dependency-ordered tickets, burns them unattended in a container, and puts a gate on the merge. If you are on macOS and your tasks are already well-shaped, Conductor is the better tool. If you are on Windows or Linux, or the expensive part of your work is deciding what to build, runcastle is.

What each one actually is

Conductor describes itself as a way to “Run parallel Claude Code, Codex, and Cursor agents in isolated workspaces on your Mac”, and its docs add OpenCode to that list. The unit of work is a workspace: “Each task gets its own workspace, branch, files, terminal, diff, and review path.” Underneath, that is git — “When you create a workspace, Conductor creates a Git worktree for that workspace and checks out a branch inside it” — with the app automating the parts you would otherwise do by hand: creating the branch and worktree, copying the local files you allow, running setup and run scripts, keeping the diff visible, tracking pull request state, and archiving the workspace when the task is done. When the work is finished, “Conductor helps you review the diff, open a pull request, merge, and archive the workspace.”

runcastle is an opinionated programming system layered on Claude Code — the IDE to Claude Code's text editor. Every feature gets a persistent session that walks six phases: ideation, spec, tickets, build, review, shipped. You get grilled on an idea until a spec and a set of tickets fall out; sandboxed AFK agents burn those tickets on the feature's branch; you test drive the result and merge. Human in the loop at only the two ends. It runs entirely on your machine — a Bun server plus a browser UI at http://localhost:4512, with no runcastle account and no hosted backend.

Side by side

Dimension Conductor runcastle
Platforms macOS. “Conductor is not available for Windows or Linux yet” — there is a waitlist. macOS, Windows, Linux.
Form factor Native Mac app, downloaded and dragged to Applications. Local Bun server plus browser UI on localhost:4512, installed with bun add -g runcastle.
Agents Claude Code, Codex, Cursor, OpenCode. Claude Code only — it is the engine runcastle drives.
Isolation One git worktree and branch per workspace, plus optional cloud workspaces on paid plans. One branch per feature; docs-only worktrees for interactive sessions; unattended builds run in a Docker or Podman container.
Where work comes from You create a workspace per task and brief the agent. Grilling produces a spec, the spec is cut into dependency-ordered tickets, and agents pull unblocked tickets.
Landing the work Review the diff, open a pull request, merge, archive the workspace. GitHub sign-in is part of setup. Test drive the branch on its own port, then click merge; gates block by default and take a recorded one-line override.
Price Free plan for local workspaces, bring your own subscriptions and keys. Pro $50/month adds cloud workspaces and multiplayer; Teams $60/user/month; Enterprise custom. Free and MIT-licensed. You still need a paid Claude plan for Claude Code.
Account Requires GitHub and at least one agent provider login; plans and cloud workspaces are account-based. No runcastle account, no hosted backend.

The real difference: where the tool starts

Conductor's model begins at the task. You have something you want done, you make a workspace for it, an agent works in that worktree, and the app carries it to a pull request. That is a genuinely hard surface to build well — the diff view, the checks, the PR state, the setup scripts that make a fresh worktree actually runnable — and Conductor is built around getting it right for several sessions at once.

runcastle begins one step earlier and ends one step later. Before any agent runs, a real Claude Code terminal argues with you until the idea is concrete, writes the decisions to docs/features/<slug>/ in your own repo, and cuts them into atomic tickets in a dependency order. That order is what makes the build parallel: everything unblocked starts at once. Then you click burn and close the tab — the tickets are burned headless inside a container, committing to the feature branch, with no session for you to sit and watch. At the other end, two gates are the only places the pipeline stops for you: the ticket review before the burn, and the merge after the test drive.

So the honest framing is not “which one runs agents better”. It is: do you want a great cockpit for sessions you are steering, or a pipeline that takes an idea and hands back a branch? Conductor invests in the first. runcastle invests in the second, and deliberately does not rebuild the chat UX at all.

Platform is a real dividing line

Conductor is Mac-only today; its install page says plainly that it “is not available for Windows or Linux yet” and offers a waitlist. That is not a knock on the product — a native Mac app is part of why it feels good — but it settles the question for a lot of people before any feature comparison starts. runcastle runs on macOS, Windows and Linux, because it is a Bun server and a browser tab rather than a desktop binary.

When Conductor is the better choice

When runcastle is the better choice

These are not mutually exclusive. Both tools are ultimately arranging git branches around Claude Code sessions on your own machine, and nothing stops a Mac user from shaping a feature in one and running errands in the other.

Sources

Everything above about Conductor was checked against its own site and docs on 10 August 2026, against release 0.79.0. Products move; if you are reading this much later, check the originals.

Claims about runcastle come from its README and CONTEXT.md, and are documented in more depth under the docs.