Parallel
One machine, many isolated Git worktrees. Status at a glance: working, blocked, waiting on you.
Pit 15 · rehearsal hall · codeorch.help
Run 100+ parallel coding agents on your machine
Source-available desktop app (ELv2). Isolated Git worktrees. Any CLI agent. Does not proxy API calls.
Claude Code, Codex, or any CLI agent — each on its own staff. We conduct. We are not a coding agent.
desktopfix onboarding crash+46 −1
cloudbilling webhooks+193
gpu-boxrefactor auth flow+394 −23
mobilespeed up cold start+33
clibump dependencies+127 −8
Cue lights · R wait · Y review · G shipELv2
Parts of the score
Parallel, any agent, diffs, MCP
Four staves. One conductor. Your keys stay in the pit — we never proxy the call.
Any agent
Claude Code, Codex, Cursor, OpenCode, Gemini, or any CLI-based tool. Switch parts whenever you want.
Diff review
Inspect the trees before you merge. Approve, comment, open the worktree in any IDE.
MCP
Hook tools the agents already speak. Drive CodeOrch from the terminal, CI, or another agent.
How it works
Ship the whole backlog, not one prompt at a time
01 · cue
Start a task
Describe what you want built. CodeOrch spins a workspace in an isolated Git worktree so every task starts on a clean branch.
02 · play
Agents do the work
Run Claude Code, Codex, or any CLI agent in parallel, each in its own workspace. No conflicts, no waiting on a single session.
03 · cut
Review and ship
Jump in when an agent needs you. Review the changes, open them in your IDE, and commit, push, and merge without leaving the app.
Parallel execution
Run 100+ agents without losing the beat
Launch agents across features, bug fixes, and refactors. The board shows which parts are playing, which are fermata, and which wait on you.
workinganalyze tab vs agent usagegen
workingPyTorch MNIST experimentsgen
reviewenterprise order flow+93 −18
reviewset up editor rules+37
Automations
Put recurring work on a schedule
Turn chores into scheduled agents: issue triage, changelog drafts, dependency bumps. They run on their own and open PRs for you to review.
codeorch automations daily-triage daily 9:00 running changelog-draft sun 11:00 ok · 2h ago dep-upgrades weekly ok · 1d ago roadmap-sync monthly ok · 3d ago
Universal compatibility
Works with any CLI agent
CodeOrch is agent-agnostic. Use Claude Code, OpenCode, Cursor, or any CLI-based coding tool.
Claude
OpenCode
Codex
Gemini
Cursor Agent
Mistral Vibe
Kimi Code
Grok CLI
Isolation
Changes stay on their own staff
Each agent runs in its own isolated Git worktree. No merge conflicts, no stepping on each other. Review and merge when you are ready.
Remote workspaces
Run workspaces anywhere
Add any machine as a host. Workspaces keep running when your laptop sleeps. Check in from wherever you are.
CLI & SDK
Spawn workspaces and agents from the terminal. Wire CodeOrch into CI, or let an agent drive it over MCP.
Open anywhere
Jump into VS Code, Cursor, Xcode, JetBrains, or any terminal. Open worktrees exactly where you need them.
Local first
Repos, worktrees, terminal output, and agent sessions stay on your machine. You choose which agents and services to connect.
Security
Private by default. You are in control.
Your code stays local by default, with explicit control over connected services.
Source available
Full source on request under Elastic License 2.0 (ELv2). Inspect, audit, contribute. No black boxes.
Local first
Repos, worktrees, and sessions stay on your machine. Cloud sync, if you turn it on, covers account metadata only.
Local control
CodeOrch runs on your machine. You pick which agents and external services to connect. BYO keys.
Not Apache
ELv2, not Apache. Do not confuse this desktop orchestrator with other projects that share a common English word.
Frequently asked
Questions from the pit
How is CodeOrch different from running Claude Code in a terminal?
CodeOrch is not a coding agent. It is an orchestrator: many CLI agents, each in an isolated Git worktree, with a board for status, diffs, and review. You spend time shipping, not waiting on one session.
I already use an IDE like Cursor. Is this for me?
Yes. Keep your editor. CodeOrch opens worktrees in VS Code, Cursor, Xcode, JetBrains, or a terminal when you want to jump in. The swarm runs beside the IDE, not instead of it.
Which AI coding agents are supported?
Any CLI-based coding tool. Claude Code, Codex, Cursor Agent, OpenCode, Gemini, and others you already run in a terminal.
How does the parallel agent system work?
Each task gets its own Git worktree and branch. Agents cannot clobber each other. You review diffs and merge when a part is ready.
Is CodeOrch free to use?
The desktop app is source-available under ELv2. You bring your own model keys; we do not meter or proxy those calls. Write hi@codeorch.help for team seats — we do not publish a rate card we do not own.
Can I use my own API keys?
Yes. BYO keys. Calls stay on your machine. CodeOrch does not proxy API calls.
Is CodeOrch open source?
Source-available under Elastic License 2.0 (ELv2), not Apache. Inspect and audit the code. Ask hi@codeorch.help for GitHub access.
What platforms does CodeOrch run on?
Download for macOS is the current public build. Write for other hosts or a CLI-only install.
Is CodeOrch just a wrapper around Claude Code?
No. It is agent-agnostic orchestration: isolated worktrees, review, automations, remote hosts, CLI, SDK, and MCP. Claude Code is one part you can seat. It is not the score.
Try CodeOrch now
Download for macOS
Or write hi@codeorch.help for the CLI, source under ELv2, and GitHub access. Paste the agent tab into Claude Code and it can install itself.
$ brew install codeorch/tap/codeorch