01
Tell the captain what you want
State the goal in one sentence, or use a ready-made template. The captain breaks the big job into small tasks, and handles splitting, assigning, and reviewing.
Your goalCaptain · opussplit · assign · review
Desktop app · macOS / Linux / WindowsOne AI as captain,
One AI as captain,
a crew of AIs doing the work
Split one task across multiple parallel AI coding sessions. Expensive models only plan, review, and make the calls; cheap models do the actual typing. Each worker AI edits only its own code copy — they can't see each other, and they can't touch your main branch until you approve the merge. Save money, and never worry about your code getting scrambled.
Free plan forever · 3 concurrent sessions · no credit card required
WebCode AI Studio — ~/work/checkout-service
i18n maintenance · CaptainClaude · opus⇄ Switch AIisolated copy
Git feat/i18n → main 2 uncommitted✓ in sync with main
› Captain Breaking down the goal: keep zh / en / ja in sync
├─ 派发 worker/i18n-zh driven by Z.ai · isolated copy
├─ 派发 worker/i18n-clean driven by Z.ai · isolated copy
└─ 派发 worker/i18n-ja driven by Z.ai · isolated copy
✓ worker/i18n-zh — filled in 12 keys · auto-approved (low risk · pure addition)
! ! worker/i18n-clean plans to delete 7 keys, 2 look dynamically referenced
→ → escalated to you: deletion would affect SessionCard.tsx
✓ worker/i18n-ja — proofread 31 entries · waiting to merge into main
› 3 auto-approved · 1 needs your confirmation · nothing pushed to remote▌
● Terminal 1+Enter · 1 · Esc
The problemRunning AIs in parallel isn't hard.
Running AIs in parallel isn't hard.
Not getting tangled up by them is.
Most people juggle three or four terminals running AIs, then spend more time figuring out "which window is this" and "what did it just change" than they actually save.
×A pile of terminal windows
- —Can't tell which window is running which task, keep switching back and forth
- —Several AIs editing the same repo at once, overwriting and conflicting with each other
- —Every step needs you watching for Enter — walk away five minutes and it stalls
- —Everything runs on the priciest official quota, even a simple rename burns money
- —Want to try another AI? Have to re-explain the whole backstory from scratch
- —It quietly committed / pushed something, and you only find out later
✓WebCode AI Studio
- ✓Sessions grouped by project, see what everyone's doing on one screen
- ✓Each AI gets its own isolated code copy — physically separated, no clashing
- ✓Low-risk changes auto-approve, only the ones that truly need you pop up
- ✓Captain runs on the official flagship, workers run on cheap providers
- ✓One click to swap to another AI CLI, the whole conversation carries over as-is
- ✓Merges only after you approve — never auto-pushes to the remote
How it works
Three steps: assign, isolate, review & merge.
02
Worker AIs stay in their own rooms
Each worker AI gets its own working directory on the same repo. The files they change are invisible to each other, and the main branch stays untouched.
Captainworker/aworker/bworker/c
03
You only nod at the important parts
Low-risk changes auto-approve; risky ones get escalated to you with the reasoning and a diff summary. Only merges to main once you approve.
worker/a ✓worker/b ⚠You decidemain
Features
A workstation designed from scratch for "running many AIs at once."
The CLIs you already use, in one interface
No new tools, no new habits. Claude Code, Codex, Gemini CLI and other sessions are grouped by project, switch on one screen — a real terminal, real process, real file tree, not a simulator.
Claude CodeClaude Code · Z.aiCodexGemini CLI
One code copy per worker, physically isolated
Not a prompt telling it "please don't mess things up" — it's separated at the filesystem level: each worker AI gets its own working directory (a Git worktree).
Swap AIs mid-conversation, no re-explaining
Model stuck, or too expensive? Hit "switch AI" and the whole conversation hands off to another CLI as close to intact as possible, and you keep going.
Claude⇄Codex⇄Gemini
Git status always in view
Current branch, how far behind main, how many files uncommitted — all shown right above the session. Have the AI read the diff before writing the commit message.
Only your calls pop up
Changes are auto-graded by risk: low-risk changes auto-approve, medium/high-risk ones come with the reasoning and changed files, queued for your word.
Skills that stick around
A workflow you finally got working can be saved as a "skill", reused instantly next time, and synced to your other machines too.
Keep an eye on it from your phonePro
Spin up a temporary phone-accessible link, check where the AI's at and where it's stuck while you're out, tap to confirm and let it keep going.
Same CLI, a cheaper provider underneath
The captain keeps its judgment on an official flagship quota, while worker AIs run Claude Code on cheap providers — cutting the cost of repetitive work by an order of magnitude.
You pick the models and tools
Bring your own model provider, MCP supported too. The captain and each worker AI can each choose their own provider and toolset.
Copy and main, side by side
Switch the file tree between the "AI's working copy" and "main" with one click, and scan through it yourself before merging.
A terminal right next to the session
Open a few more terminal tabs in the same working directory to run builds and tail logs, no need to open another terminal app and cd around.
When it asks, just answer right there
When the AI stops for confirmation, Enter, number keys, and Esc sit right beside the input box — no need to switch back to the terminal to find the cursor.
CostSame work,
Same work,
no need to pay flagship prices for all of it.
In a typical i18n maintenance task, only "how to split it" and "the final review" deserve the most expensive quota. The other ninety percent is repetitive labor — plenty handled by Claude Code running on a cheap provider.
Illustrative numbers, actual cost depends on your task, provider, and chosen models.
Pricing
Start free, upgrade once it clicks.
Free is enough to run a full end-to-end pass: 3 concurrent sessions, 1 worker AI assigned. Pro unlocks unlimited parallel sessions and workers.
Free
$0forever
- ✓Up to 3 concurrent sessions
- ✓One code copy per person, no interference
- ✓Captain assigns work (limited to 1 worker)
- ✓Multi-CLI and multi-provider management
- ✓Local skills and tool (MCP) configuration
Pro 年付
¥648/yr
- ✓无限会话
- ✓Orchestrator 并行多 worker
- ✓移动端公网隧道
- ✓技能云同步
- ✓年付约省 2 个月
Pro 月付
¥65/mo
- ✓无限会话
- ✓Orchestrator 并行多 worker
- ✓移动端公网隧道
- ✓技能云同步
Whose money pays for the model usage?
Yours. WebCode AI Studio is just the workstation — it uses your own Claude Code, Codex, Gemini CLI accounts, API keys, and providers. Usage is billed directly to your existing accounts.
Does "switch AI" lose context?
No restart needed. Switching hands off the current conversation to another CLI as completely as possible, and you keep going from there. Extremely long sessions may trim some detail, but the requirements, constraints, and gotchas carry over.
Will it commit code or push to the remote by itself?
No. Changes made by worker AIs stay only in their own code copy — they merge back to the main branch only when you click "merge"; pushing to the remote is always initiated by you.
Do I need to switch away from the CLI I already use?
No. It just manages the CLI process already installed on your machine — a real terminal, not a simulator, so your existing config, login state, and shortcuts are all still there.
Does my code get uploaded to your servers?
No. All sessions and files run locally on your machine. A temporary, closable channel is only opened when you actively turn on remote viewing from your phone.
Get startedThis afternoon's work,
This afternoon's work,
done in parallel by a crew of AIs.
Download it, log in with your existing CLI, and run your first orchestration in five minutes.
macOS·Linux·Windows