Superset vs Conductor vs Shuriken (2026)

<!-- Superset is renaming (their post, 2026-10-12, updated 2026-10-21). When the new name is out: swap "Superset" for "<New name> (formerly Superset)" in the title, description and first mention, "<New name>" elsewhere. Keep the slug. -->

We make Shuriken, so this is not a neutral page. We tried to be fair to all three; corrections welcome at office@shuriken.sh.

Quick answer#

Conductor is the Mac app with managed cloud workspaces, Multiplayer and an iOS app on Pro. Superset is the source-available Mac app (Linux experimental) with a CLI, SDK and MCP server, cheap seats and an iPhone app. Shuriken is the one with Windows and stable Linux builds, adds your own servers, and moves running chats between your accounts when one runs out.

All three run several coding agents at once, each in its own isolated copy of your repo, on your own plans and keys. On 12 October 2026 Superset announced it is changing its name (their post); we'll update the name here when it's out.

At a glance#

Read on 2026-10-22. Every Conductor and Superset cell links to that company's own page. Where a company's pages don't mention something, we say so rather than guess.

Pick Conductor if…#

• Everyone who'll use it is on a Mac.
• You want agents to keep running in a managed cloud while your laptop sleeps, and you'd rather not own a server. Conductor's cloud workspaces come with Pro; the pricing page says there's no extra fee today and usage-based pricing for cloud compute is planned.
• Live collaboration matters now. Multiplayer ships on Pro for up to five people, and Teams has an admin portal and central billing.
• You want a native iOS app today.

More in Shuriken vs Conductor.

Pick Superset if…#

• You're on a Mac (or happy with an experimental Linux AppImage).
• You want to drive your agents from scripts: a CLI, a TypeScript SDK, or an MCP server another agent can call.
• Your buyer asks for SOC 2. Superset says it completed a SOC 2 Type II audit, with the report on Enterprise.
• Price per seat matters most. $15–$20 per user is the lowest paid tier of the three.
• You like being able to read the source (ELv2).

More in Shuriken vs Superset.

Pick Shuriken if…#

• Anyone on the team uses Windows or Linux, or you want agents on a headless server you own (Linux, macOS or Windows), installed from the app with the agent CLIs set up for you.
• You run several Claude, ChatGPT or Gemini accounts and you're tired of logging in and out. Add each as a source; when one reaches your switch point (99% by default) or hits a limit, your chats move to the account with the most room, and a card in the chat says so. Claude Code chats move on their next request, subagents included. See source switching.
• You want limits to come back sooner. With session limits, Shuriken starts an idle Claude account's 5-hour window with one tiny background request, so it resets hours earlier than if it waited for you.
• You want one MCP sign-in for everything. Global MCPs work in any project, on any account and any device.
• You talk to your agents. Voice turns your speech into text on your own computer by default, and text you paste mid-sentence lands where you were.
• You use agents beyond the big three. Shuriken runs 67 of them, each with the sign-ins you already have.

Only add accounts you own and are entitled to use. Providers' terms may not allow sharing a subscription sign-in.

Where your agents actually run#

This is the question that splits the three most clearly, so it's worth a closer look.

Conductor runs agents in local workspaces on your Mac, or, on Pro and up, in Conductor's own cloud workspaces. Its routines run in the cloud too. If you want work to continue with your laptop shut and you don't want to look after a machine, that's the simplest option of the three. The trade-off is that the work runs on someone else's computer, and Conductor says usage-based pricing for cloud compute is coming.

Superset says code runs "on your machine" by default: repos, worktrees and agent sessions stay local, and its cloud sync covers account and organisation metadata only (pricing FAQ). You can add any machine as a host, so work keeps running when your laptop sleeps, and Superset's own comparison also lists cloud workspaces (comparison).

Shuriken runs agents only on computers and servers you own. A server can be Linux, macOS or Windows; the installer sets up the agent CLIs and GitHub sign-in and keeps itself updated to signed releases. Shuriken's cloud signs you in, links your devices and relays end-to-end encrypted messages. It runs no agents and stores no code. And on a MacBook, Cocaine mode keeps the machine working with the lid closed, which is exactly as dramatic as it sounds and much less illegal.

What none of them does for you#

None of the three sells model usage. You bring your Claude, ChatGPT or other plans and API keys, and the limits are the provider's. A harness can make those limits easier to live with, but it can't raise them. If limits are the problem you're solving, our guide to Claude's 5-hour limit explains how the windows really work.

How we'd test them#

Spend an afternoon, not a week:

1. Open the same repository in each and start three agents on three small tasks.
2. Close your laptop lid for ten minutes. Note which agents kept going, and where they ran.
3. Run one long Claude Code task on an account that's nearly out. Note what happens at the limit.
4. Check each one from your phone.
5. Merge one change and count the clicks.

If you want more options than these three, see the best parallel coding agent tools. If Shuriken fits, download it: the free plan needs no card.