TaskWing vs Conductor
Conductor is a Mac app for running several Claude Code and Codex sessions side by side, each in its own git worktree, with you steering each one. TaskWing is an open source command-line harness: you give one task, your agent splits it, agents build the pieces in cloud sandboxes, and a hidden exam checks the joined work before one draft pull request comes back. Pick Conductor to drive many sessions yourself today; pick TaskWing for one checked result from an open source tool.
At a glance
6 of 8 for TaskWing · 2 of 8 for Conductor
| What matters | TaskWing | Conductor |
|---|---|---|
| Open source you can read and self-host | Yes | No |
| One task split across agents for you | Yes | No |
| Checked by tests the agents never see | Yes | No |
| Agents run without your GitHub login | Yes | No |
| One pull request for the whole task | Yes | No |
| Free to use | Yes | Partly |
| Several agent runtimes today | No | Yes |
| Finished desktop app today | No | Yes |
Yes yes Partly partly No not today
The difference in one line
Conductor is a cockpit: you open many agent sessions and steer each one. TaskWing is a harness: you hand over one task, and it plans, builds in parallel, examines and returns one pull request.
Side by side
| TaskWing | Conductor | |
|---|---|---|
| What it is | Open source harness, command line | Mac app |
| Source | Open, AGPL-3.0 | Closed (no published source we could find) |
| Cost | $0; you bring your agent and sandbox accounts | Free tier for local use; paid plans for cloud workspaces |
| Agents | Claude Code today; Codex CLI or OpenCode next | Claude Code and Codex |
| Where agents run | One boat sandbox per agent; local folders to try it | Git worktrees on your Mac; cloud microVMs on paid plans |
| Who splits the work | Your lead agent, with a check that no two pieces share a file | You, by opening sessions |
| Checking the work | A hidden exam, written before work starts and run in a clean sandbox | Diffs and checks per workspace, reviewed by you |
| What comes back | One draft pull request for the whole task | A diff and pull request per workspace |
| Agent access to your GitHub login | Never; your machine pushes | Local sessions run on your Mac, with your own tools and logins |
| Maturity | Proof of concept | Shipping product |
When Conductor is the better pick
- You want a finished Mac app with a good interface, today.
- You like to steer each agent, compare approaches and pick the best diff yourself.
- You want Codex and Claude Code side by side now. TaskWing runs only Claude Code today.
When TaskWing is the better pick
- You want to give one task and get one pull request back, not manage several sessions.
- You want the result checked by tests the agents never saw, not only by your reading of each diff.
- You want agents kept away from your GitHub login, with only your machine allowed to push.
- You want to read, change and self-host the code that runs your agents.
How the work is checked
This is the biggest design difference. In Conductor, each workspace shows its diff and its checks, and you decide what to keep. In TaskWing, a separate agent session writes the checks before any work starts, you approve them, and they run in a clean sandbox on the joined branch. The workers never see the hidden tests, so they cannot write code that only passes the test. More on why in why the agents never see the tests.
Questions
Is Conductor open source?
We could not find published source code for Conductor; it is a commercial app with a free local tier and paid plans. TaskWing is open source under AGPL-3.0.
Do both run Claude Code?
Yes. Conductor runs Claude Code and Codex. TaskWing runs Claude Code today, with Codex CLI or OpenCode planned next.
Where do the agents run?
Conductor runs agents in git worktrees on your Mac, and in cloud workspaces on its paid plans. TaskWing runs each agent in its own boat sandbox, or in local folders to try the flow.
Can I use both?
Yes. They solve different steps. You might use Conductor to explore and steer sessions by hand, and TaskWing when you want a whole task split, built and examined without watching it.
Which is more mature?
Conductor. It is a finished app with a UI. TaskWing is a proof of concept with a command line, and its desktop app still runs a demo engine.
Sources
Checked on 18 November 2026. Products change fast; if something here is out of date, open an issue and we will fix it.
Try the proof of concept
TaskWing is open source (AGPL-3.0) and early. Today it runs Claude Code agents in boat sandboxes, or in local folders to try the flow.
go install github.com/josephgoksu/TaskWing/cmd/taskwing@main
taskwing run "add a /health endpoint with a test" --provider local --no-pr