MANTIS sits quietly beside you while you model in Grasshopper — reading your canvas the whole time. It stays out of your way until something silently breaks or you get stuck. Then it's already there, with the reason and the one-click fix. A partner in the room, not a box you talk to.
Most of the time you lose in Grasshopper isn't the moment you ask for help — it's the twenty minutes a definition is quietly broken and you don't know yet. It turns green, looks fine, and produces nothing — because three components upstream, a wire is feeding the wrong kind of data. Grasshopper doesn't warn you. The AI in the side-panel doesn't know, because you didn't ask. Nobody is actually watching the work with you.
Every AI tool for Grasshopper — and the old MANTIS too — works the same way: a panel you turn to, type at, and wait on. A good human assistant doesn't wait to be asked. They're in the room, glance over, and say "oh — your loft is empty because that curve isn't closed." Quiet 95% of the time; right when they speak. That's the product.
Open Rhino + Grasshopper and a calm green dot appears in the corner. Not a panel demanding attention — more like the light that says something's on and working. It reads your canvas the whole time. You never explain your file. You forget it's there.
An hour in, you rewire a cluster. What you don't notice: an output just went empty. The dot shifts — quietly — to amber. No popup, no interruption. You keep working. Maybe you'll catch it. You don't.
You're jiggling a slider, confused why nothing updates. You glance at the dot and hover. A small card: "Your panel grid went empty when you rewired Divide. Want me to fix it, or show the plan?" You click Fix. It reconnects on your actual graph — not a rebuild — and shows a receipt: "Fixed. Now produces 96 panels." Fifteen minutes saved.
At day's end you close Grasshopper. MANTIS leaves two lines: "Today: parametric façade, 96 panels, verified clean. One fix applied to the Divide cluster." Tomorrow it picks up where you left off.
Ask in plain English and MANTIS stages the graph, then places native, editable Grasshopper components on your canvas — wired and grouped, exactly as an expert would. It reads the result back and heals its own warnings. The same engine that watches is the one that builds.
Describe a design — it plans, then streams grouped native components onto the canvas.
Acts across Rhino and Grasshopper. Asks before it deletes.
A conversational answer about GH or your current canvas — explains, doesn't build.
Builds on what's there — panelization, thickness, mullions — nothing dropped.
Every competitor — Raven, Ant, Reer — runs on metered cloud AI: every look at your model is billed, so they can't afford to watch continuously. MANTIS runs on your key or a local model, where watching is free. The always-on partner is the one shape their pricing forbids them from building.
Because MANTIS is the only one actually watching, it sees what no panel can: not what you ask, but where you struggle — silently, before you'd ever ask. Opt-in and anonymous, geometry never shared, that becomes a living map of how to get unstuck. Competitors start that map at zero.
Clippy failed because it talked too much and was usually wrong. MANTIS is the opposite by design.
It speaks only when it caught something real and you're genuinely stuck — you tried, it didn't work, you paused. Dismiss it once and it backs off.
It reads the real result of your live solve, so "this is empty and here's why" is a fact it machine-checked — not a guess.
Every fix touches your actual graph and is a single Ctrl-Z. Anything destructive asks first.
Local, on your own key or a local model. Your geometry never leaves; only opt-in, anonymous, corroborated lessons do.
The plugin lives in Rhino. Your geometry never leaves unless you send it. No cloud lock-in.
Bring your own API key, or run a local model with Ollama / Claude CLI. No MANTIS subscription, ever.
It's a real GH definition — every node Ctrl-Z-able, inspectable, and yours to extend.
Free, open source, and watching your canvas inside Rhino 8 in two minutes.