Fleet 2.9.1 is out.See what's new →
FleetFleet
v2.9.1

Install Fleetctl

One command, prebuilt binary, under a minute. You will need an active Fleet trial or subscription before it can start hosted workflows.

Licensing. Fleet is a paid product after its free tier with 500 hosted starts. The binary is free to install and inspect, but new hosted workflows require a current entitlement. If you do not have a registration code yet, start free.

Let your AI install Fleet

If you use Claude Code, Codex, OpenCode, or any assistant that reads SKILL.md — install the fleet-setup skill once and then just tell it to set Fleet up. It walks you through prerequisites, the install, registration, sign-in, fleet up, and skills install.

1

Install the skill

One-time install. Uses npx skills to clone the skill from our public repo and drop it into your agent's skill directory.

terminal
$npx skills add Feral-Intelligence/agent-skills --skill fleet-setup
2

Ask your AI assistant

Open any repo you want Fleet to manage and send your agent a plain-English request. It handles the rest: OS detection, binary install, license registration prompt, project init, and picking the right first agent.

“Set up Fleet in this repo”

You still need an active trial or subscription. The skill prompts for a registration code from the dashboard and will not proceed without one.

+

Then let your AI run the fleet

Once Fleet is set up, add the fleet-manager skill. Your assistant becomes the operator of your agent team: it scopes requests into tickets, selects the right saved workflow, and follows its governed run to a reviewed, approved merge instead of writing the code itself.

terminal
$npx skills add Feral-Intelligence/agent-skills --skill fleet-manager

“Manage my fleet”

+

Then keep what the team learns

fleet-lessons is the governed Learning contract. After install, fleet skills install and the Fleet Codex plugin both include it, so the assistant loads signed policy before work and turns corrections into human-approved Lessons instead of one-off prompt edits.

terminal
$fleet skills install

“Remember this as a Lesson”

Prefer to install by hand? The three manual steps are below.

Install Fleet manually

One command. Downloads a prebuilt binary, verifies its SHA-256, and installs it to ~/.local/bin/fleet. No Go toolchain, no source build, no repository access required.

1

Install Fleet

Runs the installer at fleetctl.ai/install. It detects your OS and architecture, fetches the latest version from /releases/latest, downloads the matching prebuilt binary, verifies the SHA-256 against the published SHA256SUMS, and installs to ~/.local/bin/fleet. Supported: macOS and Linux on x86_64 and arm64. This is the official Fleetctl CLI. The npm package named fleetctl is a different product (FleetDM at fleetdm.com); Fleetctl is not published on npm, PyPI, or Homebrew.

terminal
$curl -fsSL https://fleetctl.ai/install | sh
# Verify the install
$fleet version
# → fleet 2.9.1 darwin/arm64 go1.26.1
2

Connect Fleet to your account

Fleet will refuse to start agents, the brain daemon, or the watcher until this step succeeds. Generate a registration code in your Fleet dashboard, then paste it here. Run fleet login afterward so hosted workflows and Lessons use your human identity.

terminal
# Generate a registration code in your dashboard first
$fleet admin register --url https://app.fleetctl.ai --code YOUR_REGISTRATION_CODE
$fleet login
# Confirm the license and identities
$fleet admin status
$fleet auth status
3

Bring the repository up

Run inside any git repository you want Fleet to manage. fleet up scaffolds .fleet/config.yaml if needed, starts the watcher, and starts the brain. Then install the skills that ship with this binary.

terminal
$cd /path/to/your/project
$fleet up
$fleet skills install
# See what the template actually created, then start one agent by DB name
$fleet agent list
$fleet status

What happens without a license

The binary is free to install and inspect. New hosted workflow starts require a registered, active trial or paid subscription.

License-gated commands

Any command that starts an agent, spins up a daemon, or runs a hosted workflow calls the entitlement gate first. An unregistered, expired, or exhausted trial produces a clear error with the exact command to fix it.

fleet agent start
fleet: not registered — run: fleet admin register --url <url> --license <key>
fleet brain start
fleet: not registered — run: fleet admin register --url <url> --license <key>
fleet watcher start
fleet: not registered — run: fleet admin register --url <url> --license <key>
fleet genflow worker
fleet: not registered — run: fleet admin register --url <url> --license <key>

Commands that work without a license

Diagnostic and read-only commands always work so you can install, inspect, and register the binary without a chicken-and-egg problem.

fleet version
fleet status
fleet doctor
fleet agent list
fleet admin status
fleet admin register

Verify your install

Run these commands after installation to confirm everything is wired up.

Health check

fleet doctor audits every dependency and reports which are OK, missing, or misconfigured.

terminal
$fleet doctor
# Expected output:
# [OK] tmux /usr/local/bin/tmux
# [OK] sqlite ~/.fleet/fleet.db (272 KB)
# [OK] data directory ~/.fleet
# [OK] gh cli /usr/local/bin/gh

System requirements

Fleet needs these tools on your machine before it can run.

tmux

Required

Fleet manages each agent in its own tmux session. Install via your package manager: brew install tmux (macOS) or apt install tmux (Debian/Ubuntu).

gh CLI

Recommended

The GitHub CLI (cli.github.com) is used by the brain daemon to poll PRs, reviews, CI checks, and issues. Run 'gh auth login' once to authenticate.

Active Fleet entitlement

Required

New hosted workflows require an active trial or paid subscription. Generate a registration code in the Fleet dashboard, then run 'fleet admin register'.

Need a registration code?

Start free and generate a registration code in the dashboard. Free includes 500 hosted starts. A start counts at durable first-step authorization even if the workflow later fails; internal steps and same-run retries or resumes do not add usage. Model-provider fees are separate.