A desktop pet that keeps an eye on Claude Code for you.

What the buddy shows

Every agent event becomes one word on the pet, so a glance is enough:

No more switching back to the terminal every minute to check, and no more finding out twenty minutes later that the agent stopped to ask you something.

Connect it in two clicks

Claude Code: right-click the buddy, then Claude Code › Connect. The buddy adds its hooks to ~/.claude/settings.json, keeps a backup next to the file and leaves your own hooks alone.

Codex: right-click, then Codex › Connect. The hooks go in ~/.codex/hooks.json. Codex will not run a hook it has not been shown, so type /hooks inside Codex once, review the WorkKit Buddy entry and trust it.

Disconnect in the same menu removes only the buddy's hooks.

It never reads your conversation

From each agent event the buddy reads three things: the event's name, the name of the tool being used, and, only to tell whether tests are running, the shell command. It never opens your prompts, the model's replies or your history, and none of it leaves your computer. The hooks sit in plain files you can open and read.

The full list of what the app can see is on the privacy page.

Your scripts can use it too

The buddy registers a link scheme, so a build script, a deploy or a long job can hand you a card when it finishes:

open -g "workkitbuddy://notify?title=Deploy&text=Production%20is%20live"

That is macOS. On Windows use start "" "workkitbuddy://say?text=hello".

More than an agent monitor

It is a full desktop pet as well: 23 characters or your own picture, dancing to whatever music is playing, reminders and meeting alerts delivered as cards, a focus timer and stretch breaks. Everything is free, and there is no account to make.