Every AI for Grasshopper is a box you stop and talk to. MANTIS was built on a different bet: that the real help isn't a smarter vending machine — it's a presence that watches the work with you and catches the silent break before it costs you the afternoon.
The expensive moment in Grasshopper isn't asking a question. It's the twenty minutes a definition is quietly broken — green, solving, producing nothing — because a wire upstream is feeding the wrong data. Grasshopper doesn't warn you. The AI in the panel doesn't know, because you didn't ask. Nobody is watching the work with you.
Every tool answered by being a better box: you stop, turn to a panel, type, wait, read, turn back. But a good human assistant doesn't wait to be asked. They're in the room — they glance over and say "your loft is empty because that curve isn't closed." Quiet most of the time; right when they speak.
So MANTIS watches your canvas the whole time, verifies what it sees against the live solve, and only speaks when it's caught something real and you're stuck — then fixes it on your actual graph in one undo. It runs on your machine, on your key or fully local. The reason no metered-cloud tool can follow: watching all day is free for you, and billed for them.
Presence is a quiet light, not a chat that talks first. It speaks only when it caught something real and you're genuinely stuck. Dismiss it once and it backs off.
It only asserts what it machine-checked against your live solve. "This is empty and here's why" is a fact, not a hunch. That's how it earns the right to interrupt.
Bring your own key or run fully local. Your geometry never leaves; only opt-in, anonymous, corroborated lessons do. Every fix is one undo.
MIT licensed, on GitHub. Read it, fork it, audit exactly what it sends. Trust is earned by being inspectable.
Star it on GitHub, or install it and build something today.