Install

hev kit is one binary, hev. Install it on every machine that sends or runs work. The daemon, hevd, is the same binary running as a service.

What you need

What else a machine needs depends on what it does. A laptop that only sends work to a shared host needs very little.

You want to	You need
Capture and search your traces locally	hev and Docker (Docker Desktop on macOS and Windows). Not needed with a hosted hev layer archive
Send work to another machine	hev and SSH access to that machine. Nothing else
Run dispatched work on this machine	hev, Git, a clone of each repo it works on, and Codex or Claude Code installed and logged in. On macOS and Linux, also tmux; Windows uses a scheduled task instead
See pull request links on runs	GitHub’s gh CLI, logged in. Optional

macOS and Linux

Install the released 0.6.1 formula (Homebrew on macOS or Linux):

brew tap hev/tap
brew trust hev/tap
brew install hev/tap/hev

Current Homebrew requires trusting the tap before installing its formula. The older hev/tap/kit name remains a compatibility formula.

Release archives

Download the archive for your OS and architecture from the 0.6.1 release and verify it against that release’s checksums.txt. macOS and Linux have kit_0.6.1_darwin_ARCH.tar.gz and kit_0.6.1_linux_ARCH.tar.gz for arm64 and amd64. Extract hev into a directory on your PATH.

On Windows 11 AMD64, download kit_0.6.1_windows_amd64.zip, extract hev.exe to a stable directory on your user PATH, and open a new PowerShell terminal. Run hev.exe --help to verify it. The Windows page covers Task Scheduler and dispatch targets.

Build from source

With Go 1.26 or later:

go install github.com/hev/kit/cmd/[email protected]

This puts hev in $(go env GOPATH)/bin. Add it to the PATH seen by SSH, or use path_prefix in dispatch config. See hosts.

Start the daemon

hev up starts what a machine needs, on macOS, Linux and Windows:

hev up

It runs these in Docker:

• the hev layer gateway (community edition), which stores and searches traces
• its Postgres store and a CPU embedding model
• the kit dashboard, at http://127.0.0.1:8099

It also writes ~/.hev/config.toml, sets up hevd (how depends on the system, below), and installs the bundled skills for Claude Code and Codex. Your transcripts stay on the machine. hev up needs Docker running and no key. To share one archive with your team, export a HEVLAYER_API_KEY and the archive goes to hosted hev layer instead.

Running hev up again checks the stack, refreshes bundled skills and restarts capture when the installed binary needs updating. hev down stops the containers and unloads the daemon. The archive stays in its Docker volume.

How hevd runs depends on the system:

• macOS: hev up installs and starts it under launchd.
• Linux: hev up installs and starts hevd.service under systemd’s user manager (details below).
• Windows: hev up starts the Docker stack (Docker Desktop) but doesn’t install the daemon yet. Run hev daemon install once to register the per-user \hev\hevd scheduled task; see Windows.

On Linux, hev daemon install, start, status, stop and uninstall manage the user service. On headless hosts, enable lingering for your user with loginctl enable-linger so it survives logout. Without a usable user manager, hev up reports the problem; run hev d in the foreground under your own supervisor.

Bundled skills

hev up installs version-stamped skills for existing Claude Code and Codex homes. hev skills check reports missing, stale or locally edited skills. hev skills sync repairs missing and stale copies; --force also replaces local edits. Check the diff before forcing a replacement.

To send traces to an existing Layer instead of a local one, run hev init and enter its endpoint, API key and namespace. hev config show prints the effective config with secrets redacted.

Verify

hev s

hev s reports the daemon’s last run, units indexed and last error. Then search:

hev traces search "why did the preflight fail"

Dispatch needs no daemon and no extra install. It needs a config file on the sending machine; the quickstart writes one.