v7.0 On-call preview·Local·Source-available·FSL-1.1-MIT

Your agents are
on call.
You are not.

Bugs from customers and from monitoring go to coding agents that fix them. CCC decides the small reversible things, messages you when it needs a real decision, and takes your answer from the same thread.

Email Telegram iMessage SMS WhatsApp ntfy

An illustration of the thread on your phone. The message text is what CCC sends. Example data.

How a bug moves
1 · Bugs come in
  • Customer email, SMS, WhatsApp
  • GitHub issues
  • Sentry
  • PostHog exceptions
  • A URL that must stay up
  • Your test suite
2 · Agents fix them
SHOP-12 Checkout button does nothing on Safari agent working
With one switch on, small reversible choices are decided and listed with an undo.
3 · Real decisions reach you
SHOP-12 needs a decision (#4)
keep it

One numbered message. Your reply is the agent's answer.

4 · Your customer hears back
The problem you reported is fixed: Checkout button does nothing on Safari. Reply to this message if you still see it.

Sent after your yes, on the channel they wrote in on.

localhost:8090 real capture · synthetic demo sessions
CCC v5.43 showing the session fleet and conversation reader, captured with synthetic demo sessions
The problem

11pm. The checkout button is dead.

A customer reports it. The fix is ten minutes for an agent, but only if someone starts it, answers its one question and tells the customer. That someone is you.

You ship a product with real users. You are the support desk, the on-call engineer and the product owner at once.

What happens instead

One bug. Five steps. See which reach you.

The same bug, with CCC On-call turned on. Every message below is the text CCC sends, with example data.

Does not reach you Reaches you
  1. 1

    The report arrives.

    Does not reach you

    Dana is on your customer list. She emails your support mailbox: the checkout button does nothing on Safari. The report becomes one ticket, SHOP-12, in a WatchTower queue, deduplicated.

    Dana, to your support mailbox
    The checkout button does nothing on Safari.
    SHOP-12 · filed once Checkout button does nothing on Safari
  2. 2

    An agent starts.

    Does not reach you

    A worker agent picks up the ticket and works on it in the repo you pointed it at. Nobody starts the fix by hand.

    SHOP-12 Checkout button does nothing on Safari agent working
  3. 3

    It asks you one real question.

    Reaches you

    The agent hits a choice it cannot decide. You get one numbered message on your phone, on the channel you set up. It works with no browser open.

    With the small-decisions switch on, minor reversible choices never get this far. The agent states its recommendation and how to undo it, and carries on.

    CCC to you
    SHOP-12 needs a decision (#4) Checkout button does nothing on Safari Keep the old redirect as a fallback for one release, or remove it now? Agent recommends: keep it for one release. Reply with your answer.
  4. 4

    You reply in the thread.

    Your reply

    Plain words, in the same email, Telegram, iMessage, SMS or WhatsApp thread. The reply becomes the agent's answer and the agent resumes.

    You, in the same thread
    keep it
    Sent to the agent on SHOP-12.
  5. 5

    The customer is told it is fixed.

    Reaches you: one approval

    You get one message when the fix lands. Reply yes and CCC tells Dana, on the channel she wrote in on.

    You can switch the fixed notice off.

    CCC to you
    Fixed: SHOP-12 (#7) Checkout button does nothing on Safari The click handler was bound before the button rendered. Bound it after render and added a test. Reply yes to tell Dana it is fixed, or no to skip.
    yes
    CCC to Dana
    The problem you reported is fixed: Checkout button does nothing on Safari. Reply to this message if you still see it.
2messages reached you
3words you typed: "keep it" and "yes"
0browser tabs you had to open

Counted from the example above. Your bugs will differ.

Interactive example

Try answering one message.

An illustration. The message text is what CCC sends. Example data.

This example runs in your browser. Nothing is sent or saved. The fix and its timing are scripted.

Your phone · On-call

SHOP-12 — Checkout button does nothing on Safari needs a decision (#4)

Keep the old redirect as a fallback for one release, or remove it now?
Agent recommends: keep it for one release.

Reply with your answer.

Type an answer, or tap the suggested reply.

SHOP-12 · Checkout button

  1. Reported · The bug reached the queue.
  2. Waiting · The agent needs your decision.
  3. Working · Your reply becomes its next turn.
  4. Fixed · You decide whether to tell the reporter.

The example ticket is waiting for your reply.

What it does, and where it stops

Five claims. Each with its limit.

CCC puts coding agents on call for your product. It is a preview. It is off until you turn it on.

01

Bugs come in by themselves.

Reports arrive from GitHub issues, a support mailbox, Sentry, PostHog exceptions, a URL uptime check, any command you already run (a test suite, a health script) and ccc oncall report. Each becomes one ticket in a WatchTower queue, deduplicated, and a worker agent picks it up.

The limit

Mail from someone not on your customer list waits for your go-ahead before an agent works on it. If Dana were not on your list, you would get this first:

New report from dana@example.com (#5) Checkout button does nothing on Safari Reply go to have an agent work on it, or no to drop it.
02

You hear when an agent stops or needs you.

One numbered message when an agent asks a question it cannot decide, an agent session dies mid-task, a queue stops moving, or a bug is fixed. The fixed notice is optional. It works with no browser open.

The limit

CCC must be running on a machine that stays on: your Mac, a Linux box or a small VM. Nothing is hosted for you.

03

Small reversible decisions do not reach you.

One switch. An agent that hits a minor, reversible choice states its recommendation and how to undo it, and carries on. Every such decision is listed with its undo.

The limit

Money, messages to customers, public posts and anything that cannot be undone always come to you.

04

Answer from the message.

Reply to the email, Telegram message, iMessage, SMS or WhatsApp message in plain words. The reply becomes the agent's answer and the agent resumes. The same thread also takes status, pause and resume.

The limit

WhatsApp and SMS need your own Twilio number. iMessage needs a Mac. WhatsApp only delivers free text within 24 hours of your last message to the number.

05

Agents can ask your customer.

An agent, or you, asks the person who reported the bug a question, or tells them it is fixed, on the channel they wrote in on. Their reply goes back onto the ticket and the agent resumes.

The limit

Agents never message a customer without your approval unless you turn that off. You see the wording first:

An agent wants to ask Dana (dana@example.com) about SHOP-12 (#6): Which Safari version are you on, and does it happen in a private window? Reply yes to send, no to drop, or reply with your own wording.
Channels

The alert arrives where you already are.

Pick one channel or several. Replying there answers the agent.

Email

NeedsAn address and an app password

Mail servers are detected from the address for Gmail, Outlook, iCloud, Yahoo, Fastmail, Zoho, AOL, GMX. Replies are matched by a private token in the thread.

Telegram

NeedsA bot token

CCC finds your chat after you message the bot once.

ntfy

NeedsNothing

Phone push with answer buttons.

iMessage

NeedsCCC running on a Mac signed in to Messages

SMS

NeedsYour own Twilio number

WhatsApp

NeedsYour own Twilio WhatsApp sender

WhatsApp only delivers free text within 24 hours of your last message to the number.

Your own command

NeedsAny script

For Slack, Discord or anything else you can script.

No channel needs a public URL or a webhook. CCC polls from the machine it runs on.

The hard stops

What always comes to you.

Agents do not decide these. The switch for small decisions does not cover them.

Always youMoney
Always youMessages to customers
Always youPublic posts
Always youAnything that cannot be undone
  • Agents never message a customer without your approval unless you turn that off.
  • Mail from someone not on your customer list waits for your go-ahead before an agent works on it.
  • Agents fix bugs in the repo you point them at. They open the change the way your queue is set up to: a commit or a pull request. You decide what deploys.
  • Every small decision an agent made without you is listed with its undo.
Setup

Three commands. Off until you run them.

CCC On-call is a preview. It is off until you turn it on. CCC must be running on a machine that stays on: your Mac, a Linux box or a small VM. Nothing is hosted for you.

# turn it on

# 1. turns the preview on, sends a test message
ccc oncall setup

# 2. where bugs come from
ccc oncall source add github

# 3. what is on, what is waiting for you
ccc oncall status

No CCC yet? Install it first.

# or in Settings

  • Settings > Advanced > Preview featuresTurns the preview on.
  • Settings > On-callDoes the setup: the same questions as ccc oncall setup.
  • What you need firstWhatsApp and SMS need your own Twilio number. iMessage needs a Mac. Email needs an address and an app password.

# day to day

# what is on, what is waiting for you
ccc oncall status
ccc oncall on | off

ccc oncall channel add email|telegram|ntfy|imessage|sms|whatsapp|command
ccc oncall source add github|sentry|posthog|http|command

# someone whose reports go straight to an agent
ccc oncall customer add ADDRESS

ccc oncall report "Checkout button does nothing on Safari"

# ask the reporter, or tell them it is fixed
ccc oncall ask REF "Which browser are you on?"
ccc oncall tell REF "This is fixed. Reload the page."
When you want detail

Open the session view.

Every agent, what it is doing, its transcript. You do not need it open for On-call to work. It is there when you want to look.

  • Works with the agents you already runCCC works across nine engines. Claude Code is first-class; Codex, Cursor, Antigravity, Kilo Code, Kimi Code, OpenCode, Grok, and Devin spawn and ingest with documented gaps.
  • Attaches to sessions you already haveCCC reads each engine's on-disk state as truth, so a session you started by hand in a terminal appears too. It does not own your sessions. Close the browser tab and they keep running.
  • Answer and steer in placeA session that ends on a real question is flagged. Type into any session from the browser. Dormant ones resume to receive it.
  • Queues and workersWatchTower is installed by default as CCC's queue engine. Tickets go through claim, fix, verify and close, and you can see which worker holds which ticket.
  • Past the usage limitA Claude session that hits its usage limit can keep going on a free model, then switch back when the limit resets. You choose when in Settings.
  • Runs on your machineSource-available under FSL-1.1-MIT. A stdlib-only Python server and a no-build browser UI. On-call sends the messages you configure, on the channels you set up. Anonymous telemetry is a public contract with no transcripts.
localhost:8090 · a session asking a question
The CCC session view with a session list on the left and, on the right, a Claude session asking how cancelled-stay refunds should work, with two options to pick from
Real capture, synthetic demo sessions. The same question On-call sends to your phone.
localhost:8090 · queue workers
The CCC workers view listing queue tickets being worked now, recently worked tickets, and the worker agent bound to each queue
Real capture, synthetic demo data. Worker agents and the tickets they hold.

Features that arrived earlier.

Ten selected features were released in CCC before comparable implementations merged in Orca or Paperclip, with leads from 4 days to 21 weeks.

CCC and Others timelines with ten numbered arrows in CCC release order, followed by four researched workflows with arrows ending in crosses. Dates, feature names and sources are available in the linked text version.
CCC public releases are compared with each named competitor’s code-merge dates, not its later packaged releases. The research panel’s × means no equivalent was established in the reviewed sources; it does not prove a competitor lacks the feature. Research snapshot: 12 November 2026.
Get it

Start free.

macOS, Linux, or Windows with Python 3.9+. Already use Claude Code or Codex? Your sessions are in the session view when it opens. On-call is a preview and stays off until you turn it on.

# one-line install

# macOS / Linux: downloads, installs, opens CCC in your browser
curl -fsSL https://raw.githubusercontent.com/amirfish1/claude-command-center/main/scripts/install.sh | CCC_FROM=landing bash

# Windows PowerShell
irm https://raw.githubusercontent.com/amirfish1/claude-command-center/main/scripts/install.ps1 | iex

# macOS / Linux: launch later
ccc start                           # then open http://localhost:8090

# Windows PowerShell: launch later
.\run.ps1

Asks before adding its hooks to ~/.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 install amirfish1/tap/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.

# the ccc CLI: your whole fleet from the terminal

# installed with CCC; talks to the running server
ccc sessions                        # live census: state, age, engine, repo
ccc models                          # every engine's models, efforts, cost
ccc spawn "fix the flaky test" --engine claude --model opus-5-5
ccc send queue-drain "also update the changelog"
ccc ask queue-drain "what's your status?"   # blocks for the reply

Address sessions by id or by name fragment. send is fire-and-forget (with --steer to interrupt a turn), ask waits for the reply. No browser, no hand-rolled curl.

v7.0.0 2026-11-12 · Live local CI on desktop and phone, recent results you can dismiss, one-button runner setup, and expanded On-call previews. Full changelog →

Put your agents on call.

Bugs go to agents. Real decisions come to you, as a message you can answer. A preview, off until you turn it on, running on a machine you keep on.