Why Mantis

The partner the cloud
can't afford to be.

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.

The thing they can't copy

To be in the room,
you have to watch all day.

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
And it still builds fast

One spec, one build —
not N round-trips.

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
Head to head

MANTIS vs the alternatives

MANTISCloud GH-AI
Raven · Ant · Reer
Rhino-MCP agents
Watches your canvasAlways-on, silentNo — a box you visitNo — per command
Catches silent failuresYes — verified on the solveOnly if you askOnly if you ask
Fixes in placeOne-click, one undoRe-generatesPer-action
OutputNative, wired GH definitionNative nodesNative, action-by-action
Editable afterEvery node, fullyNative nodesYes
Runs whereYour machineTheir serversYour machine
Your model / keyBYO key or fully localTheir model + meterVaries
Build speedOne call → local buildOne call, cloud computeN sequential round-trips
Self-heals warningsReads back & repairsSometimesSometimes
CostFree · open sourceSubscription (~$50/mo)Free / DIY
Reasoning shownStreamed plan, learn as it buildsVariesTool log
The honest part

No black boxes. No lock-in.

⌂

On your machine

The plugin runs inside Rhino. Geometry stays local; only your prompt and chosen context go to the model you picked.

🔑

Your provider

Claude, OpenAI, Gemini, OpenRouter — or fully offline with Ollama and the Claude CLI. Swap any time.

↺

Yours to keep

When MANTIS finishes, you own a real GH file. Extend it, version it, ship it — with or without MANTIS installed.

See the speed for yourself.

Install free and watch a 20-component definition land in one pass.