Every AI tool for Grasshopper — Raven, Ant, Reer — is a box you stop and talk to. MANTIS is the one that stays in the room: watching your canvas, catching the silent break, fixing it before you've lost the afternoon. The reason no one else does this isn't taste — it's economics. And when you do ask it to build, it builds native definitions in one fast pass.
A real partner doesn't wait to be asked — it notices the silent break (a graph that turns green and produces nothing) the moment it happens. To do that, you have to watch the canvas continuously.
Every competitor runs on metered cloud AI: every glance at your model is billed, so their unit economics forbid watching all day. MANTIS runs on your own key or a local model, where watching costs ~nothing. The always-on partner is the one product shape their pricing can't follow you into. They can clone a feature in a weekend — they can't clone their way out of their own meter.
# metered cloud — pay per glance watch the canvas all day → $$$ (so: don't) you ask → it answers → meter stops # MANTIS — your key / local model watch the canvas all day → ~free notices the silent break before you ask verifies on the live solve · one-click fix
An MCP agent works the canvas like a remote control: place a component, wait, read the result, decide the next action, repeat. Every step is a full model round-trip. A 20-component definition is 20+ sequential calls.
MANTIS emits the entire definition as one structured spec, then a deterministic builder lays it down locally. The model is called once — the wiring, layout and grouping happen at native speed on your machine.
# Rhino-MCP — agentic, per-action → add_component("Loft") …wait → add_component("Series") …wait → connect(a,b) …wait → read_canvas() …wait × 20 more round-trips # MANTIS — one spec → local build → one streamed plan + spec → deterministic builder places all 20 → reads back · heals warnings · done
| MANTIS | Cloud GH-AI Raven · Ant · Reer | Rhino-MCP agents | |
|---|---|---|---|
| Watches your canvas | Always-on, silent | No — a box you visit | No — per command |
| Catches silent failures | Yes — verified on the solve | Only if you ask | Only if you ask |
| Fixes in place | One-click, one undo | Re-generates | Per-action |
| Output | Native, wired GH definition | Native nodes | Native, action-by-action |
| Editable after | Every node, fully | Native nodes | Yes |
| Runs where | Your machine | Their servers | Your machine |
| Your model / key | BYO key or fully local | Their model + meter | Varies |
| Build speed | One call → local build | One call, cloud compute | N sequential round-trips |
| Self-heals warnings | Reads back & repairs | Sometimes | Sometimes |
| Cost | Free · open source | Subscription (~$50/mo) | Free / DIY |
| Reasoning shown | Streamed plan, learn as it builds | Varies | Tool log |
The plugin runs inside Rhino. Geometry stays local; only your prompt and chosen context go to the model you picked.
Claude, OpenAI, Gemini, OpenRouter — or fully offline with Ollama and the Claude CLI. Swap any time.
When MANTIS finishes, you own a real GH file. Extend it, version it, ship it — with or without MANTIS installed.
Install free and watch a 20-component definition land in one pass.