How long install takes

Honest read on install duration. The 30-minute fast path assumes a lot of prerequisites are already met. On a fresh machine, the real figure is one to two hours.

This page mirrors docs/INSTALL_TIME.md.

If you already have the prerequisites

If all of the following are already true on the machine you are installing on, expect 30 minutes from git clone to a green doctor.sh:

• macOS with Homebrew installed and on PATH.
• gh auth login already done; gh auth status reports a token with repo scope.
• Claude Code first-run completed: claude runs, you have signed in, and your subscription is active.
• A GitHub org or personal account you can create issues, labels, and PRs in.
• One repo cloned locally under your intended WORKSPACE_ROOT/product/.
• Python 3.11+ available (the runtime uses tomllib).

In that case the actual work is small: clone the repo, run ./bin/alfred-init.py --non-interactive --agents starter --repos my-org/my-app --slack-webhook skip, run bash deploy.sh, run bash bin/doctor.sh.

If you are setting up a fresh machine

Add 30 to 90 minutes if any of these are not done:

• Paid Claude Code subscription not set up. Going through the subscription flow plus the first claude sign-in is 15 to 30 minutes on a good day, longer if billing is being added to an org account.
• GitHub org permissions not in place. Creating an org or getting added to an existing one with repo scope (and verifying you can create issues, labels, and PRs there) typically runs 10 to 20 minutes.
• Slack workspace admin not handled. If you want Slack reports, either an incoming webhook URL or a bot token must be created. Webhook is 5 minutes; bot token (with chat:write and the channel invite) is 15 to 30 minutes if you need admin approval.
• Codex (optional engine). If you want the hybrid engine fallback, installing and signing into Codex adds 10 to 15 minutes.

A realistic fresh-machine total is 60 to 120 minutes.

One-time gates before you start

These are the things you cannot do from inside the install script, so check them first:

• You have a paid Claude Code subscription (or a Bedrock setup) and claude runs locally.
• You have a GitHub account or org where you can create issues, labels, and PRs in the repos you intend to operate against.
• You can run gh auth login and end up with a token that has repo scope.
• You have a Slack workspace where you can either add an incoming webhook or create a bot user. Skip this and pass --slack-webhook skip to the installer; Slack can be added later.
• You have one or more repos cloned locally that you want Alfred to operate against. Alfred does not clone them for you.

If any of those is “no”, expect to spend more time on that than on Alfred itself.

What happens after install

Once doctor.sh is green, the host scheduler starts the firing cadence immediately. The first Drake firing will happen within the configured interval (default every 2 hours). If you want a faster first signal, file one agent:implement issue manually and wait 20 minutes for Lucius’s next firing, or run alfred run lucius --force to trigger a firing on demand.

See also

• Install: the actual install steps.
• AI-assisted install: copy-paste prompt for Claude Code or Codex to do the install for you.
• Dry-run mode: watch a full firing lifecycle with no LLM call before turning the fleet on for real.