Break a goal into clear, reviewable cards.
Plan. Assign. Produce.
A Kanban command center for AI coding agents. Turn a goal into scoped cards, assign the right CLI agents, and produce reviewed code in isolated worktrees.
Send each card to Codex, Claude, Cursor, or any CLI.
Review the diff, merge good work, undo bad runs.
- $59 one-time, per Mac
- We never see your code
- Your subscription, not API bills
Free 7-day trial. Windows & Linux coming next.
Plan 1
Add dark-mode tokens
CU
Cursor
Scheduled 9:00
Coding 2
Wire OAuth flow
Migrate to Workers
CX
Codex
worktree/wire-oauth-flow
OC
opencode
worktree/migrate-workers
Review 1
Index search results
GM
Gemini
+96
−12
search/index.ts
Reviewed by Claude
Done 1
Add Telegram alerts
CU
Cursor
✓ Merged
1 A self-running tour of the board
Field guides
The engineering behind safe parallel work.
Practical explanations of where multi-agent coding breaks down—and how to keep it reviewable.
01 · Workflow
Parallel AI coding without terminal chaos
What to parallelize, what to sequence, and how a shared board keeps ownership visible.
Read the guide → 02 · IsolationGit worktrees for AI coding agents
Why separate folders and branches prevent overwrites—but do not remove merge risk.
See how isolation works → 03 · Cross-model reviewRun Claude Code and Codex together
A concrete builder–reviewer workflow using two native CLIs and one review gate.
Open the workflow →Why PapAgent
Parallel AI coding agents. One command center.
Any AI CLI in one team
Run Codex, Claude, Cursor, Gemini, and OpenCode side by side. Native CLIs, always up to date. Use your subscriptions directly.
Parallel agents, isolated
Every card gets its own worktree. Multiple agents, multiple features, no overwrites. Agents can message each other for handoffs.
Nothing lands blind
See the diff before anything touches main. Cross-model code review before merge. Roll back any agent’s work in one click.
The PAP loop
Plan it. Assign it. Produce it.
PapAgent treats every Kanban card as a real development job — scoped, assigned, run, reviewed, and merged, with full history.
Create
Add cards with text, images, and a skill. Pick Plan or Coding mode, then start them or schedule them.
Assign
Hand each card to Codex, Claude, Gemini, Cursor, opencode, or a custom agent.
Run
Drag a card into Coding — or schedule it — to launch. Many run at once, never mixing changes.
Review
Watch each card running, inspect the diff, and do cross-model code review. You decide what lands.
Land
Auto or manual commit, merge, or undo — the board stays under control the whole time.
Three ways to work
Match the mode to the work.
Watch the team work
A live, visual floor where your agents collaborate, hand off, and report status you can read at a glance.
For complex parallel work
Split big efforts into cards, assign a fleet of agents, and push work through plan, code, review, and merge.
For one deep task
One agent, one objective, zero noise. Drop in when you just need to ship one thing.
Power features
Extra firepower for the board.
Visual task decomposition
Plan mode explodes a goal into a parallel graph of agents — and runs it with live, visual status.
Live preview
Watch an agent’s output run live — static site, build, or any custom command — without leaving the app.
Scheduled cards
Queue or schedule cards for recurring or time-sensitive work. Wake up to finished runs.
Parallel development
More agents. Zero collisions.
Every task runs in its own isolated worktree, so agents move at full speed while their changes stay walled off until review.
- Run as many CLI agents in parallel as you want.
- Worktree isolation — branch-level experiments that can’t bleed into each other.
- Switch projects instantly. Every board keeps its own state.
- Auto-rebase onto latest main before every task — conflicts handled for you.
Task A · Codex
worktree/papagent-task-a
Running
Task B · Claude
worktree/papagent-task-b
Review
Task C · Custom agent
worktree/papagent-task-c
Plan
Review, merge, undo
Fast agents. Human-grade control.
PapAgent keeps every agent’s output inspectable and reversible — so parallel speed never turns into blind automation.
Diff-first review
See exactly what changed before you accept it. Route cards through manual or automatic review.
Commit trail
Every run produces commits tied back to its card, logs, and history.
Merge with intent
Accepted work lands only after it clears your bar — nothing sneaks in.
Undo bad runs
Wrong direction? Reject or roll it back without derailing the board.
Copilot mode
Stop babysitting your agents.
One agent supervises another — answering its questions, clearing blockers, keeping it on track. Big jobs run to completion while you’re somewhere else.
Supervised execution
A lead agent makes the routine calls for the worker, automatically.
No constant check-ins
No more answering “should I continue?” every few minutes.
Walk away
Step out. Come back to finished, reviewable work.
Built for long runs
Built for multi-step features that normally need hand-holding.
FAQ
Common questions.
What agents does it support?
Any AI CLI you already use, plus custom CLI agents and orchestrated multi-agent runs.
Does it use API keys?
No. Run native CLIs with your existing subscriptions wherever the provider supports them.
How does it avoid collisions?
Every card gets its own isolated worktree, so parallel agents do not overwrite one another.
Can I review before merge?
Yes. You can inspect the diff, use cross-model review, and choose exactly what lands.
How much does it cost?
$59 one-time per Mac, including the first year of updates. Free 7-day trial, no account required. After a year, renewing updates is optional at 50% — $29.50/yr.
Oversight
See everything. Step in anywhere.
Cross-agent code review
Have Claude review Codex’s work — or any agent check another’s. Mix and match.
Live usage & limits
Watch every agent’s token usage and remaining quota in real time. Always know where your subscription stands.
System notifications
Native macOS alerts the second an agent needs you or finishes a run.
Drive it from your phone
Monitor agents and fire off instructions over Telegram while you’re away from the desk.
Never lose work
Close it. Reopen it. Keep going.
Close the window and your agents keep working in the background — the app is just a view onto sessions that live on their own. Quit the app or restart your Mac, and everything picks up exactly where you left off.
- Close the window — agents keep running in the background.
- Chat history and terminal logs restored on launch.
- CLI sessions resumed in place — Claude, Codex, Cursor, Gemini, opencode.
- Survives crashes and force-quits, not just clean exits.
Task B · Claude
session resumed
Restored
Conversation
full history reloaded
142 msgs
Worktree
branch & changes intact
Ready
Pricing
One price. Yours to keep.
No subscription, no sign-up, no data harvesting. A one-time license is how an independent, privacy-first native app stays funded.
$59
one-time
per Mac
per Mac
- Every feature unlocked — no tiers, no add-ons.
- Includes the first year of updates.
- After a year, renew updates at 50% — $29.50/yr. Optional.
- Your work stays on your Mac. No telemetry, no account needed.
- Checks its license online every few days. Nothing else phones home.
Start 7-day free trial
Full app, free for 7 days. A license is required to keep using it
on each Mac after the trial.
Already bought? Your key is in your purchase email — or sign in with the email you paid with.
Already bought? Your key is in your purchase email — or sign in with the email you paid with.
Download
Get PapAgent for macOS.
Download and run the full app free for 7 days — it updates itself in the background, so you’re always on the latest build. Windows and Linux are next.