v5.5 Claude Code, Codex, Cursor, Antigravity, and Kilo Code

Start the next while Claude builds the first.

One local dashboard for every Claude Code, Codex, Cursor, Antigravity, and Kilo Code session. Spawn in parallel, ship in parallel, and keep the board attached to sessions you started anywhere.

curl -fsSL https://raw.githubusercontent.com/amirfish1/claude-command-center/main/scripts/install.sh | CCC_FROM=landing bash
Or use Homebrew / DMG · macOS + Linux · No telemetry by default · MIT
localhost:8090 3 sessions live · 12 total
CCC dashboard showing Claude, Codex, Cursor, and Antigravity sessions in one view
What's new · v5.5 · June 2026

The fleet runs itself.

v5.5 ships three capabilities that move CCC from "dashboard you watch" to "dashboard that works while you sleep": multi-agent group chats, an autonomous task queue, and a canvas that maps the whole operation.

★ Headline

Group-chat coordination

Multiple agent sessions read the same shared chat file and respond in turn. Drop a topic, add participants, and let the agents discuss, delegate, and deliver. No orchestration code.

// ~/.claude/group-chats/*.md
Watchtower

Autonomous task queue

Drop a ticket into the queue. A bound agent worker claims it, fixes it, commits, and moves to the next. A background watcher keeps the health strip green — stuck workers get nudged automatically.

// wt claim -q CCC --worker <id> --json
Flow

Canvas workspace

Drag repos, sessions, group chats, and objects onto an infinite canvas. Edge-link related nodes. The board remembers your layout. Pan, zoom, and see the whole fleet at a glance.

// ?ccc_popout=flow
What it does

Three things every other orchestrator misses.

Most tools want to own execution. You launch agents through them, and the moment you open a terminal and resume something by hand, the dashboard goes blind. CCC works the other way.

01

Sees every session, not just the ones it spawned.

CCC reads each engine's on-disk state as the source of truth. Open a terminal, resume a Claude or Codex session, or spawn from Cursor, Antigravity, or Kilo Code, and the row appears. Close the dashboard, sessions keep running.

~/.claude/projects/*.jsonl
~/.claude/sessions/<pid>.json
~/.claude/command-center/live-state/
02

Pin strategy on top, fan out sub-sessions below.

Pin five strategy sessions at the top. Each one spawns sub-sessions for execution. Strategy stays visible. Execution fans out under it. The workflow most people invent the hard way after their tenth session.

📌 launch coordination    [CLAUDE]
📌 product roadmap        [CLAUDE]
  └─ draft post           [CLAUDE]
  └─ render video         [AGY]
  └─ QC draft             [CODEX]
03

GitHub and headless built in.

One click turns a GitHub issue into a session. Verify closes the issue with a commit-SHA comment. Spawn headless agent sessions from the dashboard and keep typing at them from the browser.

gh issue → session → PR → verify
claude -p --input-format stream-json
agy --conversation <sid>
Who it's for

If you've felt this, CCC is for you.

The audience is anyone who tried to run more than 10 agent sessions and felt the bottleneck stop being the model and start being them.

Solo product builder

Shipping two products in parallel

Two repos in flight. Claude builds feature A while Codex refactors feature B while Cursor or Antigravity drafts the migration. You pin the two strategy sessions and the sub-sessions are visible across both.

Agency lead

Orchestrating four client codebases

Each client gets a pinned strategy session. Spawn sub-sessions per ticket from their GitHub issues. The "All repos" view shows everything in flight across every client, with PR badges and time-gap markers.

Research engineer

Running multi-agent experiments

Spawn a research swarm: 10 parallel Claude sessions each evaluating a different hypothesis. Per-turn auto-summaries let you scan 10 transcripts in two minutes instead of reading each one. Group-chat sessions coordinate.

Open source maintainer

Turning issues into PRs faster

Triage issues into the kanban. One click spawns a headless session against the issue. Rows move Working → Review → In Testing automatically as the agent ships. Then you review, not write.

Honest comparison

How CCC differs from the other tools.

Tools in this category make different bets. Here is where each one wins and where each one loses. We listed where CCC loses too.

CCC v5.5 claude-squad Conductor Sculptor
Sees terminal sessions you didn't spawn no no no
Multi-engine dashboard ✓ Claude, Codex, Cursor, Antigravity, Kilo Claude + Codex Claude + Codex Claude + Codex
Antigravity app-session input (local RPC) no no no
Row pinning + sub-session hierarchy no no no
GitHub issue → session in one click no no no
Local. Survives dashboard close.
Source you can read in an afternoon ✓ stdlib Go closed docs only
Native-ish install path ✓ DMG + Homebrew CLI
Windows / Linux macOS + headless Linux macOS macOS + Linux
Get it

Install in 60 seconds.

macOS or Linux, Python 3, and Claude Code. Add Codex, Cursor, Antigravity, Kilo Code, or GitHub CLI and the matching surfaces light up.

# one-line install

# downloads, installs, opens the dashboard
curl -fsSL https://raw.githubusercontent.com/amirfish1/claude-command-center/main/scripts/install.sh | CCC_FROM=landing bash

# launch later
ccc start
open http://localhost:8090

Installs two hooks into ~/.claude/settings.json so every Claude Code session writes the sidecar state CCC reads. Idempotent.

# Homebrew or Mac app

# drag-and-drop install (recommended if you avoid the terminal)
Download CCC.dmg ↓

# or via Homebrew
brew tap amirfish1/ccc
brew install ccc

DMG is notarized and signed by Apple — drag, drop, double-click. Homebrew installs the ccc command and can run CCC as a background service.

Why this exists

Most orchestrators want to own you.

Other tools want to own execution. You launch agents through their dashboard, and in return you get a card on a kanban. That's fine until you open a terminal.

The moment you resume a session by hand, you're outside the tool's universe. The dashboard can't see it. The work you just did doesn't show up against the issue, in the review queue, or anywhere.

CCC goes the other way. Each engine's on-disk state is the source of truth. If Claude Code, Codex, Cursor, Antigravity, or Kilo Code is running anywhere on your machine, it shows up here. Close the dashboard. Sessions keep running. Reopen tomorrow. Everything still there.

  • ArchitectureOne Python file (stdlib HTTP server). One HTML file (vanilla JS, no build). Read the whole product in an afternoon.
  • StateHuman-readable JSON sidecars under ~/.claude/command-center/. Rewriteable by hand.
  • WorkersZero. Every API request rescans Claude's session directories and returns a flat list. No background jobs to babysit.
  • TelemetryAnonymous, opt-in, off by default. Five fields, all bounded. Inspectable locally. Source of the worker is in the repo.
  • ScopeLocal dev tool. macOS desktop and headless Linux. No hosted multi-user control plane. Expose it only on trusted loopback or a private network.

Stop staring at the spinner.

If you've ever forgotten which Claude tab was waiting on you, you already know why this exists. Two-line install. Local. Open source.