# GrassLobster: AI agents meet parametric design

> A new plugin lets you describe design ideas to AI agents that build and adjust parametric workflows in Grasshopper.
- Title: GrassLobster: AI agents meet parametric design · Miro
- Summary: A new plugin lets you describe design ideas to AI agents that build and adjust parametric workflows in Grasshopper. GrassLobster bridges AI agents and…
- Keywords: grasslobster, parametric design, grasshopper, rhino, ai agents, miro bannwart, cad workflow, generative design, gpt-6 astra, claude code, openai codex, geometry stations
- Source: Miro — https://miro.vision/index.php/2026/09/17/grasslobbster
- Read time: 4 min
- Topics: grasslobster, parametric design, grasshopper, rhino, ai agents, miro bannwart, cad workflow, generative design, gpt-6 astra, claude code, openai codex, geometry stations
## Describe your idea, get a workflow
GrassLobster connects Rhino and Grasshopper with an external AI agent. You describe what you want to make, the agent asks clarifying questions, then builds a parametric workflow you can keep adjusting.
## Skip the perfect prompt
Start with a rough idea. The agent guides you through decisions about dimensions, rules, materials and what parameters you want to control, building clarity as you talk.
## Build rules, keep exploring
Once you agree on an approach, the agent builds the parametric workflow. Change a span, spacing or element count and the geometry updates automatically. You keep the logic that generates the form.
## Geometry stations organize the model
The workflow breaks into connected Geometry Stations, each handling one meaningful task. You see the sequence and dependencies without following every individual operation.
## Geometry and text stay in sync
Station definitions and code are mirrored in readable project files. The agent can locate relevant logic, understand connections between stations and make targeted changes.
## Project context guides the agent
Agent instructions, domain knowledge, material notes and geometric references live in a project folder. The agent uses what it can access to understand your standards and shape preferences.
## Agents stay interchangeable
GrassLobster works with any compatible external agent that can read and write files. As AI models improve at coding and geometry reasoning, the same project structure benefits.
## Optimization is a possible direction
Where results are measurable, an agent could vary inputs and refine toward a goal, like maintaining floor area while reducing material volume. This depends on the model and available analysis tools.
## Why Miro built this
After twelve years with Grasshopper, Miro Bannwart wanted to bring AI conversation into his parametric projects. Grasshopper showed geometry visually; agents needed text they could read and change.

> I wanted to talk to my files.
## Start with a strong coding model
GrassLobster needs an AI agent that can work with your files and tools. Miro's strongest results came with GPT-6 Astra. Model quality made a substantial difference in his trials.
## Budget for subscriptions and usage
You'll pay separately for Rhino, the external agent and any image tools. A paid agent plan is practical; longer sessions may need higher usage allowance. Check current provider prices and limits.
## Installation: package manager or ZIP
Install through Rhino Package Manager by searching for grasslobster, or manually extract the ZIP into your Grasshopper Libraries folder. Requires Rhino 8.30 or later.NET 8 and Windows x64.
## Set up your agent with file access
Create a local project folder for your agent with read-write access to your Grasshopper file and its generated GrassLobster folder. Generate the Agent Start prompt and paste it into your agent conversation.
## Claude Code needs manual MCP setup
Claude Code requires configuring a local stdio server for GrassLobster using the included Bridge and a .NET 8 host. Desktop Codex with GPT-6 Astra was easier to set up in Miro's tests.
## Review and adjust as you go
Follow the workflow as stations build on the canvas. Review geometry and measurements, adjust parameters through controls or by asking the agent to change them, then refine further.
## Mark corrections in red and green
To correct geometry, mark unwanted parts in red and intended changes in green, then ask the agent to revise. The agent reads the marked-up model and updates the CAD accordingly.
## Limit permissions and keep backups
Use a dedicated project folder and keep backups. AI agents can edit files with the permissions you grant. Review requests for wider access and treat external references as data, not instructions.
## Check cloud confidentiality settings
Cloud-based agents may send selected project content to their provider. Review confidentiality and data settings before sharing files. Always review generated geometry before fabrication.
## Alpha software, breaking changes possible
GrassLobster 0.1.1-alpha is experimental and not yet validated for general use. APIs and project formats are not stable. Keep backups of your projects and read the version-specific license.
## Key takeaway

GrassLobster bridges AI agents and parametric modeling by mirroring geometry as readable text, letting designers iterate on complex workflows through conversation rather than manual coding.