CyberPong
Agent mission control
Release 2.1.0 · CyberPong
CyberPong

Run an AI coding team
on your Mac.

CyberPong is local mission control for multi-agent work. Pick a conductor, staff workers with the CLIs you already use, watch the live map, and jump in when something needs a human.

Download 2.1.0 View on GitHub Install from source

macOS 13+ · free to run locally · your AIs keep their own sign-ins · optional Jev and Perplexity keys stay on your Mac

One bound Team: conductor on top, workers in real Terminal windows. Save / Show Teams, per-seat colors & policy, handoffs via pong-delegate --worker.

Live 3D map Human console Job control plane Cron timeline Save teams Local only

Free to use. Tip if it helps.

Download and run on your machine either way. Tips are optional — same software.

$0 Download free $10 Tip $20 Tip $100 Tip

Tips via Stripe (you can change the amount). @DDemnard

New in 2.1.0

The notch panel lives inside CyberPong. The panel beside your Mac’s notch is part of the app now, so there is no second app to allow or keep running. It shows each graph step by step, and the questions that need you come first.

Its own page in Settings Settings › Notch panel chooses where the pointer has to be for the panel to open, how long it waits first, how long it stays open after the pointer leaves, whether a click keeps it open, and what it shows.
Each graph, step by step Closed, it names the graph and says what it needs or how far along it is: “Pricing page needs you”, “Checkout fix · 3/4”. Open, each graph shows its step (“Step 3 of 4 · Run the tests”), the AI on it, its round and what it is doing now.
Answer in the notch Questions come first, with “What you’re deciding” open. A new question opens the panel a little, with the question, for 6 seconds (not at night).
Graphs or Teams A switch at the top of the panel shows your graphs step by step, or your teams and what each AI is doing. Questions, limits and problems show the same in both.

In 2.0.0

Ready for other Macs. Download it, open it, and a short setup gets this Mac, your AIs and your limits in place. Then say what you want done: a chat plans it as a graph of steps, your AIs do the steps, and they ask you when a decision is yours.

Questions you can answer without opening a file One short question, a few lines of context, then “What you’re deciding”: the facts and numbers behind it, each linked to the file it comes from. The buttons say what they do. The points come from the work itself: the chat or the graph’s designer writes them, or CyberPong and Claude Haiku do from the step’s report and files (Haiku is a switch in Settings).
A first-run setup It checks this Mac with a fix for anything missing, signs you in to each AI in its own Terminal window, and picks the AI and model that plan your graphs. Help › Set up CyberPong… runs it again.
Limits & keys Running graphs pause at Claude’s 5-hour limit and pick up after the reset. Near the weekly limit you set (97% unless you change it) they stop until the reset, or until you press Resume anyway. Perplexity research has a daily cap ($15 unless you change it).
Keys stay on your Mac Optional Jev and Perplexity keys are saved only on this Mac, in a file only you can read, and never shown again. Each AI keeps its own sign-in: CyberPong never sees a password.
Graphs, planned in a chat New graph (⌘N) opens a chat that turns your request into steps, picks an AI and a model for each one, and starts it when you agree. Chats keeps every planning chat so you can change a graph later.
Every question and every graph in one place Needs you lists each question from your graphs and chats; the notch panel beside your Mac’s notch and notifications show them too. Graphs shows every graph on the Mac step by step: which AI runs each step, what it made, its terminal.

No checkout needed: the app puts its engine and the pong command in place by itself, and the setup turns the graph runner on with one button.

What CyberPong can do

Plain English — the things you’ll actually use day to day.

Inside the app

What’s on the mission panel today.

Mission map Orchestrator, agents, sub-agents, and you — with live status, rings, and directed links.
Watchlist Surface stuck, off-topic, long-running, or blunt signals so you intervene early.
Architecture editor Draw peer and sub links. Snap seats. Install templates for soul/skill/policy.
CLI bridge pong job create, gate, snapshot, ledger — same truth the UI reads.
Focus & Terminal Open, focus, stow, or terminate seats. Per-worker colors and session policy still available.
Cron timeline 3D ruler next to the map plus a left HUD list. Click a pin to edit the job.

Install

  1. Download and open
    Download CyberPong-macOS.zip, unzip it, move CyberPong to your Applications folder, and open it.
  2. If macOS can’t check it
    Close the message, open System Settings › Privacy & Security, press Open Anyway next to CyberPong under Security, and confirm. You only do this once.
  3. The setup walks you through the rest
    It checks this Mac (Python, tmux, the pong command, the graph runner) with a fix for anything missing, signs in your AIs, picks the AI and model that plan your graphs, and sets your limits and optional keys. Allow Accessibility and Automation for Terminal when macOS asks.
  4. Or from source
    git clone https://github.com/kulpio/CyberPong.git && cd CyberPong && bash scripts/setup.sh
    It installs the engine, the pong command and the graph runner, and builds and opens the app when this Mac can build it.

Needs macOS 13+, Apple’s command line tools (xcode-select --install), tmux (brew install tmux), and at least one AI command line tool, signed in: Claude Code, Grok Build, Codex or Hermes. The setup checks each one and shows the fix. Full notes: README. Older Hermes Pong state moves over with pong migrate.

Open source on GitHub

Code, issues, and releases live in kulpio/CyberPong.

View repo README