Launched this week
Minnie
Give your AI agent a body — a Mac pet that does real work
10 followers
Give your AI agent a body — a Mac pet that does real work
10 followers
Your AI agent already has a brain. Minnie gives it a body — a floating 3D pet for your Mac that does real work by driving the agent you already use: Claude Code, Gemini, Codex, or any CLI. Before anything risky runs, a cartoon speech cloud shows the exact command — deny or allow, by voice or click. Her 8 moods mirror the agent live. She talks, listens, and remembers. No telemetry, secrets in Keychain, zero extra AI cost.
Love the body for the agent idea, and the speech cloud approve/deny is a really thoughtful safety layer. One thing I'd love: let Minnie pipe allowed commands through a quick dry-run preview on long-running scripts, showing the diff of files she'll touch before I tap allow. Would save me from that "uh, what did I just approve" feeling.
@nazls86733 Thank you — that "uh, what did I just approve" feeling is exactly the enemy Minnie exists to kill.
Honest take on your idea: for file edits, very doable — the agent already streams structured edit previews, so showing the diff inside the consent cloud before you tap Allow is a natural upgrade. It's going on the roadmap today.
For arbitrary shell scripts, a true dry-run is trickier (running the script is the side effect), but I'm exploring a middle ground: surfacing the paths a command will touch when derivable — and the activity timeline already logs every step after the fact.
Quick question back: would diff-on-edits cover most of your "what did I approve" moments, or is it the shell scripts that bite you?
And if you want to kick the tires when it ships — join the waitlist here: https://pratik-singh-web.github.io/minnie-landing/ (30-second form, locks in the $19 founder price). 🐾
The speech cloud showing the exact command before it runs is genuinely clever — finally a way to let an agent loose without white-knuckling it. She also feels surprisingly alive on the desktop, not just a gimmick.
@helindiyar5o3i This genuinely made my day — "without white-knuckling it" is exactly the feeling I built the cloud for. It started as a safety feature and became the thing that made me finally trust my own agent to just run. Glad it lands the same way for you.
And thank you for the "not a gimmick" — that was the hardest line to walk. The rule I held to: she only reacts when something real happens (her moods are the agent's live state, not random cuteness).
If you want it the day it ships: https://pratik-singh-web.github.io/minnie-landing/ 🐾
Love the concept of a little 3D buddy driving your CLI agent and showing you the command before it runs. One thing that would seal it for me: let Minnie run as a menu bar icon with a quick "what did my agent just do" history pop-up, so I can glance back at the last few actions without opening the full window when I'm heads down on other work.
@asiyekahra13396 Ha — you just described half of Minnie's design brief. She already lives only in the menu bar (paw icon, no Dock), and there's a full "What has Minnie done?" timeline — searchable, every command/edit/result logged. So the heads-down use case is exactly the target.
What you're really asking for is the glance layer: the last 3 actions inline in the menu-bar dropdown, so you don't even open the timeline. That's a genuinely good addition and it's small — going on the roadmap. (There's also a status chip above her head that hover-expands to show the running/finished tool rows, if you'd rather glance there.)
Question back: last-3-actions in the dropdown, or would you want it to nudge you — a subtle badge when something finished while you were away?
Kick the tires when it ships: https://pratik-singh-web.github.io/minnie-landing/ — 30-second waitlist, locks in the $19 founder price. 🐾
honestly the speech cloud approval thing is such a smart idea, no more doomscrolling through logs to catch what the agent is about to do
@talha6u9u Right?? Logs are past-tense — you read what already happened. The cloud is present-tense: it comes to you, before the command runs, and waits. And it shows the command verbatim, never a friendly "tidying up some files" summary — that paraphrase is exactly where trust leaks. Really appreciate you getting it. 🐾
If you want it the day it ships: https://pratik-singh-web.github.io/minnie-landing/ 🐾