Conductor vs Paradev: which parallel Claude Code runner should you use?

Conductor and Paradev (Paradev was known as Superset until its October 2026 rebrand) both let you run parallel coding agents in isolated git worktrees. Both start free with your own API keys. And both have spent the last several months racing toward the same capabilities: cloud workspaces, mobile remote control, scheduled automations, and multi-model routing.

The gap between them has narrowed. What once read as "simple CLI runner vs heavy platform" is now more nuanced. Conductor has a full GUI, an MCP server, Routines for GitHub-triggered automation, and an iOS app for remote control. Paradev shipped plugins, a workspace issue tracker, and PR review threads. Choosing between them now comes down to pricing structure, scale ceiling, and how much model diversity your team needs.

Conductor: polished Mac app, flat-rate pricing, GitHub-native automation

Conductor started as a CLI tool that wrapped Claude Code in worktree-isolated parallel sessions. It is now a fully-featured Mac app: local-first, bring your own API keys, but with a GUI workspace manager, cloud sync for remote access, and an iOS app added in v0.90.0 (October 2, 2026). The latest release, v0.92.0 (October 23), added full remote control from iPhone: browsing and creating workspaces, reading agent chat, picking models, and stopping agents from your phone.

Two additions since July stand out. Routines (v0.85.0, September 9) let you schedule agent runs on a cron schedule or trigger them automatically when a GitHub event fires: a PR opened, a label applied, or a new issue from a specific author or branch. This turns Conductor into a lightweight CI layer: review or draft work starts the moment the event fires, without manual dispatch. The MCP server (v0.82.0, August 20) goes further: external agents can manage Conductor workspaces programmatically, creating and stopping sessions as part of a larger automated pipeline. Conductor Cloud, the monitoring and team layer, has continued to develop since its July launch.

The pricing structure is notable. Conductor Pro is $50 per Mac per month, not per seat. A solo developer or a pair working on the same machine pays $50 total. Conductor Teams is $60 per user per month, adding centralized billing, admin controls, and team-wide live collaboration for any size team. The per-machine Pro tier is unusual in this category and makes sense for developers who want cloud access and automation without paying per seat.

Where Conductor stands out:

• GitHub-triggered Routines. Agents start automatically on PR open, label change, issue creation, or other GitHub events, filterable by branch, author, or label. No manual dispatch required.
• MCP server for programmatic control. External agents can create, monitor, and stop Conductor workspaces via the Conductor MCP. Useful for teams building automation layers on top of Conductor's dispatch.
• Flat per-machine Pro pricing. $50 per Mac covers the whole machine regardless of how many people share it. Simpler to budget for solo developers and pairs than per-seat models.
• Polished Mac interface. Workspace management, diff viewing, model picker with saved loadouts, and Stacks (named workspace collections). The CLI origin is gone from the primary experience.

Where Conductor adds friction:

• macOS only. Conductor runs on Mac. Windows users are on a waitlist with no announced timeline. Teams with developers on Windows cannot deploy Conductor uniformly.
• Teams pricing jumps steeply. The step from Pro ($50/Mac) to Teams ($60/user/month) is significant if your team needs live collaboration and admin controls. It is not a linear per-seat addition to Pro but a separate tier.
• Not open source. Conductor is proprietary. There is no community repo or self-hosted option.

Paradev: 100+ agents, ten coding tools, $11.5M behind it

Paradev rebranded from Superset on October 26, 2026, branding only: product, accounts, settings, and billing are unchanged (the superset CLI command still works). The name change resolved conflicts with Apache Superset and unrelated projects. Version 2.0.0 shipped with the rebrand, backed by a $11.5M seed round led by Union Square Ventures across 35 investors. As of the announcement, the platform had 145,000 developers across 180 countries and 1.85 million agent workspaces created.

The scale number is Paradev's main differentiator: 100+ simultaneous agents per session. For workflows that parallelize at volume (test generation across a large codebase, multi-path prototyping, documentation sweeps) Paradev's ceiling is higher than any other tool in this category. The model support list is also genuinely broader: Claude Code, Codex, OpenCode, Cursor Agent, Copilot CLI, Mistral Vibe, Kimi Code, Grok CLI, Gemini, and any CLI-based agent. Paradev also added a bring-your-own-provider layer (OpenRouter, AWS Bedrock, Vertex, Vercel AI Gateway) for teams with negotiated model contracts.

Two notable features shipped around the rebrand. The plugins system (October 25) lets users install a plugin once and have it activate across all agents and all coding tools on every machine. For teams standardizing tooling across multiple models, this is a meaningful consistency layer. Version 2.0.2 (October 29) added PR review threads in the diff pane, so code review happens inside the workspace without switching to GitHub.

Where Paradev stands out:

• Scale. 100+ parallel agents per session is the practical ceiling in this category. For high-volume batch work, no other tool reaches it.
• Ten-tool model routing. Claude Code, Codex, Cursor Agent, OpenCode, Gemini, and six more. Teams that experiment across models or assign agents to specific tools can manage all of them from one interface.
• Cheaper per-seat for multi-person teams. Paradev Pro is $16 per user per month. Five developers on Pro pay $80 total; the equivalent Conductor Teams tier is $300. The gap widens with team size.
• Plugins across agents. A plugin installs once and activates across Claude Code, Codex, Cursor, and all other supported agents on every developer's machine , a consistency gain for teams running multiple coding tools.

Where Paradev adds friction:

• macOS only (Linux experimental). Like Conductor, Paradev does not run on Windows. Linux support is available but marked experimental. Cross-platform teams still face a gap on Windows.
• Cloud workspaces are Team tier and in beta. Remote cloud execution (agents running without your machine) is gated to the $54 per user per month Team tier and is still in beta. Free and Pro tiers run locally only.
• More configuration surface. Supporting ten coding tools and a plugin ecosystem means more decisions at setup. Teams standardized on Claude Code alone may not need that surface.

Which to pick

For most small teams (three to five developers, primarily Claude Code), the practical difference has narrowed. Both tools handle parallel worktree isolation well. The decision comes down to pricing structure and whether you need scale above roughly ten agents or multi-model routing. For larger teams or teams mixing models, Paradev's per-seat economics and multi-tool support pull ahead. For teams that want GitHub-triggered automation in a tight Mac interface, Conductor's Routines and MCP server fit better. A broader look at where parallel runners, skills frameworks, and lifecycle tools compare is in the agentic coding patterns that actually scale guide. The Claude Code orchestrators roundup covers both in context alongside the full landscape.

The layer both leave open

Conductor and Paradev are parallel dispatch layers. They solve for throughput: more agents running simultaneously, across more models, triggered on more events. What neither provides is the structure above the dispatch layer: deciding what to build before agents start, a design gate that verifies direction before implementation runs, and enforced review that checks agent output before it merges. A parallel runner dispatches work efficiently; a development lifecycle decides what work gets dispatched, in what order, and whether it is ready to ship.

That structure layer is where defract operates: a gated lifecycle (story, design, architecture, implementation, review, release) running on top of Claude Code, with a visual design stage and specialist review agents that check each other's work before you sign off. It is not a replacement for Conductor or Paradev. It sits at a different point in the stack, before parallel dispatch, deciding what the agents build and enforcing that the right things happen before code ships. For teams where throughput is not the bottleneck, where the constraint is coordination, correctness, and keeping the wrong things from shipping, why AI coding agents need hard stage boundaries covers why that process layer matters.