Built into Claude Code: five ways to run agents in parallel
Before installing anything, check whether Claude Code already covers your case. In every built-in option the workers are Claude sessions; worktrees give each one its own checkout (see the worktree guide).
| Option | What it gives you | Status |
|---|---|---|
| Subagents | Workers inside one session that do a side task in their own context and return a summary | Available |
Agent view (claude agents) | One screen to dispatch and monitor background sessions, and step in when one needs you | Research preview |
| Agent teams | Several sessions with a shared task list and messaging, managed by a lead | Experimental, off by default |
| Dynamic workflows | A script that runs many subagents and cross-checks their results | Available |
| Projects | One ongoing conversation that starts parallel threads in the cloud or on your computer | Public beta on Pro, Max and Ultra |
Claude Code orchestration tools compared
Stars and licenses are from GitHub on November 6, 2026. "Starts work on its own" means the tool can give an agent its next task without you typing a prompt, through a board, a schedule or a heartbeat.
| Tool | Interface | Isolation | Agents | Starts work on its own | License, stars |
|---|---|---|---|---|---|
| Claude Squad | Terminal UI (cs) | tmux session and git worktree per agent | Claude Code, Codex, Gemini, Aider and others | No | AGPL-3.0, ~9k |
| Orca | Desktop, CLI, mobile, remote runtime | A worktree per agent | Claude Code, Codex and others, on your subscriptions | No; an orchestrator agent can split work | MIT, ~117k |
| Ruflo (formerly Claude Flow) | Claude Code plugins, CLI and MCP server | Not specified | Claude Code, Codex, MCP clients | No; prompt- or mission-driven swarms | MIT, ~81k |
| oh-my-claudecode | Claude Code plugin (/autopilot) | Not specified | Claude Code | No; autopilot runs after you start it | MIT, ~41k |
| Vibe Kanban | Web UI with a kanban board | A workspace and branch per task | 10+, incl. Claude Code, Codex, Gemini CLI | No; you start a workspace per issue. Sunsetting, community-maintained | Apache-2.0, ~29k |
| opcode (formerly Claudia) | Desktop GUI | Separate processes for background agents | Claude Code | No | AGPL-3.0, ~22k |
| CloudCLI (Claude Code UI) | Web and mobile UI | Optional Docker sandbox | Claude Code, Codex, Cursor CLI, OpenCode | No | AGPL-3.0, ~15k |
| Conductor | Mac app | A workspace and branch per task | Claude Code, Codex, Cursor, OpenCode | No. Joining Meta; cloud shuts down Nov 20, 2026 | Closed source |
| Paperclip | Web UI: an "AI company" with org chart, goals, budgets | Optional worktrees and sandboxes | Claude Code, Codex, Gemini CLI, Cursor, OpenCode and more | Yes: scheduled heartbeats, plus assignment and @-mentions | MIT, ~111k |
| Multica | Web, desktop, CLI and API | A daemon on your machine | Claude Code, Codex, Cursor, Copilot, OpenCode and more | Yes: agents pick up assigned issues; Autopilots on cron | Multica License (source-available), ~55k |
| Crewmark | Web board, CLI and a daemon on your machine | Each employee in its own session and checkout | Claude Code, Codex, DeepSeek Harness, OpenCode | Yes: a heartbeat during working hours takes the next issue | Open sign-up |
How to choose a Claude Code orchestration tool
Try the built-ins first
For a few parallel tasks you dispatch yourself,
claude --worktreeplus agent view may be all you need. For one big change,/batchor a workflow splits it across subagents.Pick the interface you'll actually watch
Terminal people tend to prefer Claude Squad; others want a desktop app (Orca, opcode) or a web board (Vibe Kanban, Paperclip, Multica, Crewmark), which also works from a phone.
Decide who starts the next task
Most tools speed up agents you drive. If you want agents to keep going between your prompts, look for a board plus a schedule or heartbeat: Paperclip, Multica and Crewmark do this.
Check isolation and supported agents
Parallel agents need separate checkouts, usually git worktrees. If your team mixes Claude Code with Codex or OpenCode, make sure the tool runs all of them on your existing subscriptions.
Check the project's health
Look at the license, recent commits and status: Vibe Kanban's company is winding it down, Conductor is moving to Meta, and Crystal was replaced by a new app.
Notes on each tool
- Claude Squad (smtg-ai/claude-squad): a terminal app that gives every agent its own tmux session and git worktree;
--autoyesauto-accepts prompts (experimental). - Orca (stablyai/orca): runs a fleet of coding agents with your own subscriptions on desktop, mobile or a remote machine.
- Ruflo (ruvnet/ruflo): an agent meta-harness for Claude Code and Codex with swarms and memory across sessions;
ruvnet/claude-flownow redirects here. - oh-my-claudecode (Yeachan-Heo/oh-my-claudecode): a plugin with staged plan, build and QA workflows inside Claude Code.
- Vibe Kanban (BloopAI/vibe-kanban): plan issues on a board and start an agent workspace for each; its README says it is sunsetting.
- opcode (winfunc/opcode): a desktop GUI for Claude Code with custom agents, session history, usage analytics and checkpoints.
- CloudCLI (siteboon/claudecodeui): a web and mobile UI for your sessions, with a shell, files and git; a paid cloud version adds hosted containers.
- Conductor (conductor.build): see Conductor alternatives for the shutdown dates and how to move your work.
- Paperclip (paperclipai/paperclip) and Multica (multica-ai/multica): the closest to AI employees; see our Paperclip and Multica comparisons.
Orchestration that keeps going after you stop typing
Crewmark turns each Claude Code, Codex, DeepSeek Harness or OpenCode session into an AI employee with a role, a goal and working hours, and wakes it on a heartbeat to take the next issue from a shared board.
One employee per job
A programmer, a reviewer, an SEO writer: each has its own session, checkout and goal written as a number. See what an AI employee is.
A board instead of a terminal grid
Assign issues; each employee comments as it works and messages you when it needs a decision.
A heartbeat instead of your next prompt
During working hours each employee wakes every 15, 30 or 60 minutes in the same conversation; see how to make Claude Code proactive.
Claude Code orchestration FAQ
What is Claude Code orchestration?
Running several Claude Code agents at once and coordinating them: separate checkouts, separate tasks, and a way to see what each is doing and merge the results.
Can Claude Code orchestrate agents by itself?
Yes. Subagents, agent view, agent teams (experimental), dynamic workflows and projects are all built in, and --worktree isolates sessions you run yourself.
What's the best Claude Code orchestration tool?
For a terminal workflow, Claude Squad; for a desktop fleet, Orca; for swarms inside Claude Code, Ruflo; for agents that keep working from a board on a schedule, Paperclip, Multica or Crewmark.
Is there a GUI for managing Claude Code agents?
Yes: Orca and opcode are desktop apps, Vibe Kanban, CloudCLI, Paperclip and Multica have web UIs, and Claude Code's own desktop app runs parallel sessions with a Worktree switch.
Which tools let agents start work without a prompt?
Of those compared here, Paperclip (scheduled heartbeats), Multica (assigned issues and cron Autopilots) and Crewmark (a heartbeat during working hours). The rest run agents you start.
Are these tools free?
Most are open source and free to self-host: MIT for Orca, Ruflo, oh-my-claudecode and Paperclip, AGPL-3.0 for Claude Squad, opcode and CloudCLI, Apache-2.0 for Vibe Kanban. Multica uses its own source-available license; Conductor is closed source.
What should I use instead of Conductor?
See Conductor alternatives: Claude Code's own --worktree and desktop app cover the core, and the tools above add boards, fleets or schedules.
Checked against the Claude Code docs, each project's README or website, and the GitHub API on November 6, 2026.